KR20200050433A - Method and apparatus for performing processing for image using neural network - Google Patents

Abstract

The present invention relates to a decoding method, a decoding device, an encoding method, and an encoding device of a video. Disclosed are a method for performing processing for an image by using a neural network, and a device thereof. A block to which the neural network is applied is generated by using the neural network in accordance with properties of a transform unit, and a reconstructed block for the transform unit is generated by using the generated block. The method for generating the reconstructed block can replace a process using an existing in-loop filter.

Description

Method and apparatus for performing image processing using a neural network {METHOD AND APPARATUS FOR PERFORMING PROCESSING FOR IMAGE USING NEURAL NETWORK}

The following embodiments relate to a video decoding method, a decoding apparatus, a coding method, and a coding apparatus, and related to a method and apparatus for processing an image using a neural network.

Through the continuous development of the information and communication industry, broadcast services having high definition (HD) resolution have spread worldwide. Through this spread, many users have become accustomed to high-resolution and high-definition images and / or videos.

In order to satisfy users' demand for high image quality, many organizations are accelerating the development of next-generation imaging devices. Users' interest in High Definition TV (HDTV) and Full HD (FHD) TVs, as well as Ultra High Definition (UHD) TVs with 4 times higher resolution than FHD TVs This has increased, and according to the increase in interest, an image encoding / decoding technique for an image having a higher resolution and image quality is required.

In image encoding / decoding, filtering may be applied to the generated image. Filtering is mainly performed on a boundary of a specified unit, and changes values of pixels of the unit's boundary.

One embodiment may provide a method and apparatus for performing processing on a conversion unit.

One embodiment may provide a method and apparatus for performing the selected processing according to the attribute of the conversion unit.

One embodiment may provide a method and apparatus for performing processing using a neural network for a transformation unit.

In one embodiment, using the neural network according to the size of the transformation unit generating a block to which the neural network is applied; And generating a reconstructed block for the transform unit using the generated block.

A method and apparatus for performing processing for a conversion unit is provided.

A method and apparatus for performing selected processing according to the properties of a conversion unit are provided.

A method and apparatus for performing processing using a neural network for a transformation unit is provided.

1 is a block diagram showing a configuration according to an embodiment of an encoding apparatus to which the present invention is applied.
2 is a block diagram showing a configuration according to an embodiment of a decoding apparatus to which the present invention is applied.
FIG. 3 is a diagram schematically showing an image segmentation structure when encoding and decoding an image.
FIG. 4 is a diagram illustrating a form of a prediction unit PU that the coding unit CU may include.
FIG. 5 is a diagram illustrating a form of a transform unit (TU) that may be included in a coding unit (CU).
6 illustrates division of a block according to an example.
7 is a diagram for explaining an embodiment of an intra prediction process.
8 is a diagram for describing a reference sample used in the intra prediction process.
9 is a diagram for explaining an embodiment of an inter prediction process.
10 shows spatial candidates according to an example.
11 shows an additional order of motion information of spatial candidates according to an example to a merge list.
12 illustrates a process of transformation and quantization according to an example.
13 shows diagonal scanning according to an example.
14 shows horizontal scanning according to an example.
15 illustrates vertical scanning according to an example.
16 is a structural diagram of an encoding apparatus according to an embodiment.
17 is a structural diagram of a decoding apparatus according to an embodiment.
18 shows the amount of error according to the size of TU division and TU according to an example.
19 shows a neural network according to an embodiment.
20 shows PSNR according to the size of TU according to an example.
21 is a structural diagram of an encoding device according to an embodiment.
22 is a structural diagram of a decoding apparatus according to an embodiment.
23 is a flowchart of operation of an in-loop network according to an embodiment.
24 shows a result of dividing a CTU into CUs according to an example.
25 illustrates input and output of an in-loop network according to the size of a TU according to an embodiment.
26 shows an example of setting a TU as an input of an in-loop network for a luma component when a 4: 2: 0 color format according to an example is used.
27 shows an example of setting a TU as an input of an in-loop network for a chroma component when a 4: 2: 0 color format according to an example is used.
28 shows the size of TU according to QT segmentation, BT segmentation, and TT segmentation according to an example.
29 illustrates a pre-processing process for input to an in-loop network according to an example.
30 shows a process of generating feature maps according to an example.
31 shows an in-loop separable network according to a TU size using a pre-processing network according to an example.
32 illustrates a method of reconstructing the output of a separable network according to an example as a reconstructed image.
33 illustrates the operation of a shared network according to an embodiment.

The present invention can be applied to various changes and can have various embodiments, and specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

For detailed description of exemplary embodiments described below, reference is made to the accompanying drawings that illustrate specific embodiments as examples. These embodiments are described in detail enough to enable those skilled in the art to practice the embodiments. It should be understood that the various embodiments are different, but need not be mutually exclusive. For example, certain shapes, structures, and properties described herein may be implemented in other embodiments without departing from the spirit and scope of the invention in relation to one embodiment. In addition, it should be understood that the location or placement of individual components within each disclosed embodiment can be changed without departing from the spirit and scope of the embodiment. Therefore, the following detailed description is not intended to be taken in a limiting sense, and the scope of exemplary embodiments, if appropriately described, is limited only by the appended claims, along with all ranges equivalent to those claimed.

In the drawings, similar reference numerals refer to the same or similar functions throughout several aspects. The shape and size of elements in the drawings may be exaggerated for a clearer explanation.

In the present invention, terms such as first and second may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from other components. For example, the first component may be referred to as a second component without departing from the scope of the present invention, and similarly, the second component may be referred to as a first component. The term and / or may include a combination of a plurality of related described items or any one of a plurality of related described items.

When a component is said to be "connected" or "connected" to another component, the two components may be directly connected to or connected to each other, but the above 2 It should be understood that other components may exist in the middle of the dog components. On the other hand, when it is mentioned that a component is "directly connected" or "directly connected" to another component, it should be understood that no other component exists in the middle of the two components. something to do.

Components shown in the embodiments of the present invention are shown independently to indicate different characteristic functions, and do not mean that each component is composed of separate hardware or one software component unit. That is, each component is included as listed for each component for convenience of description. At least two components of each component are combined to form one component, or one component is divided into a plurality of components to function. It can be carried out and the integrated and separated embodiments of each of these components are included in the scope of the present invention without departing from the essence of the present invention.

In addition, in the exemplary embodiments, the description that "includes" a specific configuration does not exclude configurations other than the specific configuration described above, and additional configurations may be used to implement the exemplary embodiments or the technical spirit of the exemplary embodiments. It means that it can be included in the scope.

The terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present invention, terms such as “include” or “have” are intended to indicate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, and that one or more other features are present. It should be understood that the existence or addition possibilities of fields or numbers, steps, operations, components, parts or combinations thereof are not excluded in advance. That is, in the present invention, description of "including" a specific configuration does not exclude configurations other than the corresponding configuration, and means that additional configurations may also be included in the scope of the present invention or the technical spirit of the present invention.

Some of the components of the present invention are not essential components for performing essential functions in the present invention, but may be optional components for improving performance. The present invention can be implemented by including only the essential components in implementing the essence of the present invention, excluding components used for performance improvement. A structure including only essential components excluding optional components used for performance improvement is also included in the scope of the present invention.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings in order to enable those skilled in the art to easily implement the embodiments. In describing the embodiments, when it is determined that a detailed description of a related known configuration or function may obscure the subject matter of the present specification, the detailed description will be omitted. In addition, the same reference numerals are used for the same components in the drawings, and duplicate descriptions for the same components are omitted.

Hereinafter, an image may mean one picture constituting a video, or may represent the video itself. For example, "encoding and / or decoding of an image" may mean "encoding and / or decoding of a video", and "encoding and / or decoding of one of the images constituting the video". It might be.

Hereinafter, the terms “video” and “motion picture (s)” may be used in the same sense, and may be used interchangeably.

Hereinafter, the target image may be a target image to be encoded and / or a target image to be decoded. Also, the target image may be an input image input to the encoding apparatus, or an input image input to the decoding apparatus. Also, the target image may be a current image that is a current encoding and / or decoding target. For example, the terms “target image” and “current image” may be used interchangeably, and may be used interchangeably.

Hereinafter, the terms “image”, “picture”, “frame” and “screen” may be used in the same sense, and may be used interchangeably.

Hereinafter, the target block may be an encoding target block that is an encoding target and / or a decoding target block that is an encoding target. Also, the target block may be a current block that is a target of current encoding and / or decoding. For example, the terms “target block” and “current block” may be used interchangeably, and may be used interchangeably.

Hereinafter, the terms “block” and “unit” may be used in the same sense, and may be used interchangeably. Or "block" may indicate a specific unit.

In the following, the terms "region" and "segment" may be used interchangeably.

Hereinafter, a specific signal may be a signal indicating a specific block. For example, the original signal may be a signal representing a target block. The prediction signal may be a signal representing a prediction block. The residual signal may be a signal representing a residual block.

In embodiments, each of the specified information, data, flag, index and element, attribute, etc. may have a value. The value "0" of information, data, flags, indexes, elements, and attributes may represent a logical false or a first predefined value. In other words, the values "0", false, logical false and first predefined values can be used interchangeably. The value "1" of information, data, flags, indexes, elements, and attributes may represent a logical true or a second predefined value. In other words, the values "1", true, logical true and second predefined values can be used interchangeably.

When a variable such as i or j is used to represent a row, column, or index, the value of i may be an integer greater than or equal to 0, or an integer greater than or equal to 1. In other words, in embodiments, rows, columns, indexes, and the like may be counted from 0 and counted from 1.

In the following, terms used in the embodiments are described.

Encoder: An encoder may mean a device that performs encoding. In other words, the encoder may mean an encoding device.

Decoder: A decoder may mean a device that performs decoding. In other words, the decoder may mean a decoding device.

Unit: The unit may indicate a unit of encoding and / or decoding of an image. The terms "unit" and "block" may be used interchangeably and may be used interchangeably.

-The unit may be an MxN array of samples. M and N may each be positive integers. A unit can often mean an arrangement of two-dimensional samples.

-In encoding and decoding an image, the unit may be an area generated by segmentation of one image. That is, the unit may be a specific area in one image. One image may be divided into a plurality of units. Alternatively, the unit may refer to the divided portion when one image is divided into subdivided portions and encoding or decoding of the divided portion is performed.

-In encoding and decoding an image, predefined processing for the unit may be performed according to the type of unit.

-Depending on the function, the type of the unit is a macro unit (Macro Unit), a coding unit (Coding Unit; CU), a prediction unit (Prediction Unit; PU), a residual unit (Residual Unit) and a transformation unit (Transform Unit; TU), etc. Can be classified as Or, depending on the function, the unit is a block, a macroblock (Macroblock), a coding tree unit (Coding Tree Unit), a coding tree block (Coding Tree Block), a coding unit (Coding Unit), a coding block (Coding Block), a prediction unit (Prediction Unit), a prediction block (Prediction Block), a residual unit (Residual Unit), a residual block (Residual Block), a transformation unit (Transform Unit) and a transform block (Transform Block). For example, the target unit may be at least one of a CU, a PU, a residual unit, and a TU that are targets of encoding and / or decoding.

-The unit may refer to information including a luma component block, a corresponding chroma component block, and a syntax element for each block to refer to the block.

-The size and shape of the unit can vary. Further, the unit may have various sizes and various shapes. In particular, the shape of the unit may include geometric shapes that can be expressed in two dimensions, such as rectangle, trapezoid, triangle, and pentagon, as well as square.

In addition, the unit information may include at least one of a unit type, a size of a unit, a depth of a unit, a coding order of a unit, and a decoding order of a unit. For example, the type of unit may indicate one of CU, PU, residual unit and TU.

-One unit can be further divided into sub-units having a smaller size than the unit.

Depth: Depth may refer to the degree of division of a unit. In addition, the depth of the unit may indicate the level at which the unit exists when the unit (s) is expressed as a tree structure.

-The unit division information may include a depth related to the depth of the unit. Depth may indicate the number and / or degree of division of the unit.

-In the tree structure, the depth of the root node is the shallowest, and the depth of the leaf node can be regarded as the deepest. The root node may be the highest node. The leaf node may be the lowest node.

-One unit may be divided into a plurality of sub-units hierarchically while having depth information based on a tree structure. In other words, the unit and sub-units created by the division of the unit may correspond to a node and child nodes of the node, respectively. Each divided sub-unit may have a depth. Since the depth indicates the number and / or degree of division of the unit, the division information of the sub-unit may include information about the size of the sub-unit.

-In the tree structure, the uppermost node may correspond to the first unit that is not split. The uppermost node may be referred to as the root node. Also, the highest node may have a minimum depth value. At this time, the highest node may have a depth of level 0.

-A node having a depth of level 1 may represent a unit generated as the first unit is divided once. Nodes with a depth of level 2 may represent units created as the first unit is split twice.

-A node having a depth of level n may represent a unit generated as the first unit is divided n times.

-The leaf node may be the lowest node, or a node that cannot be further divided. The depth of the leaf node may be the maximum level. For example, the predefined value of the maximum level may be 3.

-QT depth may indicate depth for quad division. BT depth may indicate the depth for binary division. The TT depth may indicate the depth for ternary division.

Sample: A sample may be a base unit constituting a block. The sample may be represented as values from 0 to 2 ^Bd- 1 according to bit depth (Bd).

-The sample may be a pixel or pixel value.

-Hereinafter, the terms "pixel", "pixel" and "sample" may be used in the same sense, and may be used interchangeably.

Coding Tree Unit (CTU): A CTU may be composed of one luma component (Y) coding tree block and two chroma component (Cb, Cr) coding tree blocks related to the luma component coding tree block. have. Also, the CTU may mean that the above blocks and syntax elements for each block of the above blocks are included.

-Each coding tree unit, such as a quad tree (QT), a binary tree (BT) and a ternary tree (TT), etc., to construct sub-units such as a coding unit, a prediction unit, and a transformation unit It may be divided using one or more division methods. A quad tree may mean a quarternary tree. Further, each coding tree unit may be divided using a multi-type tree (MTT) using one or more splitting schemes.

-CTU can be used as a term to refer to a pixel block that is a processing unit in a process of decoding and encoding an image, such as in segmentation of an input image.

Coding Tree Block (CTB): The coding tree block may be used as a term to refer to any one of a Y coding tree block, a Cb coding tree block, and a Cr coding tree block.

Neighbor block: A neighbor block may mean a block adjacent to a target block. The neighboring block may mean a reconstructed neighboring block.

-Hereinafter, the terms "neighbor block" and "adjacent block" may be used with the same meaning, and may be used interchangeably.

-The neighboring block may mean a reconstructed neighboring block.

Spatial neighbor block: A spatial neighbor block may be a block spatially adjacent to a target block. The neighboring block may include a spatial neighboring block.

-The target block and the spatial neighboring block may be included in the target picture.

-The spatial neighboring block may mean a block having a boundary with a target block or a block located within a predetermined distance from the target block.

-The spatial neighboring block may mean a block adjacent to the vertex of the target block. Here, the block adjacent to the vertex of the target block may be a block vertically adjacent to a neighboring block horizontally adjacent to the target block or a block horizontally adjacent to a neighboring block vertically adjacent to the target block.

Temporal neighbor block: The temporal neighbor block may be a block temporally adjacent to the target block. The neighboring block may include a temporal neighboring block.

-The temporal neighboring block may include a co-located block (col block).

-The call block may be a block in a co-located picture (col picture) that has already been reconstructed. The position in the call picture of the call block may correspond to the position in the target picture of the target block. Alternatively, the position of the call block in the call picture may be the same as the position of the target block in the target picture. The call picture may be a picture included in the reference picture list.

-The temporal neighboring block may be a block temporally adjacent to the spatial neighboring block of the target block.

Prediction mode: The prediction mode may be information indicating a mode that is encoded and / or decoded for intra prediction or a mode that is encoded and / or decoded for inter prediction.

Prediction unit: The prediction unit may mean a base unit for prediction such as inter prediction, intra prediction, inter compensation, intra compensation, and motion compensation.

-One prediction unit may be divided into a plurality of partitions or smaller prediction units having a smaller size. The plurality of partitions may also be a base unit in performing prediction or compensation. The partition created by the division of the prediction unit may also be a prediction unit.

Prediction unit partition: The prediction unit partition may refer to a form in which the prediction unit is partitioned.

Reconstructed neighboring unit: A reconstructed neighboring unit may be a unit that has been already decoded and reconstructed in a neighbor of a target unit.

-The reconstructed neighbor unit may be a spatial neighbor unit or a temporal neighbor unit to the target unit.

-The reconstructed spatial neighboring unit may be a unit in the target picture and a unit already reconstructed through encoding and / or decoding.

-The reconstructed temporal neighbor unit may be a unit in a reference image and a unit that has already been reconstructed through encoding and / or decoding. The position in the reference image of the reconstructed temporal neighboring unit may be the same as the position in the target picture of the target unit, or may correspond to the position in the target picture of the target unit. Alternatively, the reconstructed temporal neighboring unit may be a neighboring block of a corresponding block in a reference image. Here, the position in the reference image of the corresponding block may correspond to the position of the target block in the target image. Here, the correspondence of the positions of the blocks may mean that the positions of the blocks are the same, may mean that one block is included in another block, and one block occupies a specific position of another block Can mean

Parameter set: The parameter set may correspond to header information among structures in a bitstream.

-The parameter set is at least one of a video parameter set (VPS), a sequence parameter set (SPS), a picture parameter set (PPS), and an adaptation parameter set (APS). It can contain one.

In addition, the parameter set may include tile group, slice header information, and tile header information. The tile group may mean a group including a plurality of tiles. Also, the meaning of the tile group may be the same as the meaning of the slice.

Rate-distortion optimization: The encoding device uses a combination of a size of a coding unit, a prediction mode, a size of a prediction unit, motion information, and a size of a transformation unit to provide high encoding efficiency. Distortion optimization can be used.

-The rate-distortion optimization method may calculate the rate-distortion cost of each combination in order to select the optimal combination among the combinations. The rate-distortion cost can be calculated using the formula "D + λ * R". In general, a combination in which the rate-distortion cost by the formula "D + λ * R" is minimized can be selected as the optimal combination in the rate-distortion optimization method.

-D may indicate distortion. D may be the mean square error of the difference values between the original transform coefficients and the reconstructed transform coefficients in the transform unit.

-R can represent the rate. R may indicate a bit rate using related context information.

-λ may represent the Lagrangian multiplier. R may include coding parameters such as prediction mode, motion information, and coded block flags, as well as bits generated by encoding transform coefficients.

-The encoding apparatus may perform processes such as inter prediction, intra prediction, transformation, quantization, entropy encoding, inverse quantization, and / or inverse transformation in order to calculate accurate D and R. These processes can greatly increase the complexity of the encoding device.

Bitstream: A bitstream may refer to a sequence of bits containing encoded image information.

Parameter set: The parameter set may correspond to header information among structures in a bitstream.

Parsing: Parsing may refer to entropy decoding of a bitstream to determine the value of a syntax element. Alternatively, parsing may mean entropy decoding itself.

Symbol: It may mean at least one of syntax elements, coding parameters, and transform coefficients of a unit to be encoded and / or a unit to be decoded. Also, the symbol may mean an object of entropy encoding or a result of entropy decoding.

Reference picture: A reference picture may mean an image referenced by a unit for inter prediction or motion compensation. Alternatively, the reference picture may be an image including a reference unit referenced by a target unit for inter prediction or motion compensation.

Hereinafter, the terms "reference picture" and "reference picture" may be used with the same meaning, and may be used interchangeably.

Reference picture list: The reference picture list may be a list including one or more reference pictures used for inter prediction or motion compensation.

-The types of the reference picture list are List Combined (LC), List 0 (List 0; L0), List 1 (List 1; L1), List 2 (List 2; L2), and List 3 (List 3; L3) ).

-One or more reference picture lists may be used for inter prediction.

Inter prediction indicator: The inter prediction indicator may indicate the direction of inter prediction for the target unit. The inter prediction may be one of unidirectional prediction and bidirectional prediction. Alternatively, the inter prediction indicator may indicate the number of reference pictures used when generating the prediction unit of the target unit. Alternatively, the inter prediction indicator may mean the number of prediction blocks used for inter prediction or motion compensation for the target unit.

Prediction list utilization flag: The prediction list utilization flag may indicate whether a prediction unit is generated using at least one reference picture in a specific reference picture list.

-The inter prediction indicator can be derived using the prediction list utilization flag. Conversely, a prediction list utilization flag may be derived using an inter prediction indicator. For example, indicating that the prediction list utilization flag is 0, which is the first value, may indicate that a prediction block is not generated by using a reference picture in a reference picture list for a target unit. The prediction list utilization flag indicating the second value of 1 may indicate that the prediction unit is generated using the reference picture list for the target unit.

Reference picture index: The reference picture index may be an index indicating a specific reference picture in a reference picture list.

Picture order count (POC): The POC of a picture may indicate the display order of the picture.

Motion Vector (MV): The motion vector may be a two-dimensional vector used in inter prediction or motion compensation. The motion vector may mean an offset between a target image and a reference image.

-For example, MV may be expressed in the form of (mv _x , mv _y ). mv _x may represent a horizontal component and mv _y may represent a vertical component.

Search range: The search range may be a two-dimensional area in which MV is searched during inter prediction. For example, the size of the search area may be MxN. M and N may each be positive integers.

Motion vector candidate: When a motion vector is predicted, the motion vector candidate may mean a motion vector of a prediction candidate block or a prediction candidate block.

-The motion vector candidate may be included in the motion vector candidate list.

Motion vector candidate list: A motion vector candidate list may mean a list constructed using one or more motion vector candidates.

Motion vector candidate index: The motion vector candidate index may mean an indicator indicating a motion vector candidate in a motion vector candidate list. Alternatively, the motion vector candidate index may be an index of a motion vector predictor.

Motion information: motion information includes motion picture, reference picture index and inter prediction indicator, as well as reference picture list information, reference picture, motion vector candidate, motion vector candidate index, merge candidate and merge index, etc. It may mean information including at least one of.

Merge candidate list: A merge candidate list may mean a list constructed by using one or more merge candidates.

Merge candidate: The merge candidate may mean a spatial merge candidate, a temporal merge candidate, a combined merge candidate, a combined bi-prediction merge candidate, a zero merge candidate, and the like. The merge candidate may include an inter prediction indicator, and may include motion information such as a reference picture index for each list, a motion vector, a prediction list utilization flag, and an inter prediction indicator.

Merge index: The merge index may be an indicator indicating a merge candidate in the merge candidate list.

-The merge index may indicate a reconstructed unit that derived a merge candidate among reconstructed units spatially adjacent to the target unit and reconstructed units temporally adjacent to the target unit.

-The merge index may indicate at least one of motion information of the merge candidate.

Transform unit: The transform unit may be a basic unit in residual signal encoding and / or residual signal decoding, such as transformation, inverse transformation, quantization, inverse quantization, transformation coefficient encoding, and transformation coefficient decoding. One conversion unit may be divided into a plurality of sub-transformation units having a smaller size. Here, the transform may include at least one of a primary transform and a secondary transform, and the inverse transform may include at least one of a primary inverse transform and a secondary inverse transform.

Scaling: Scaling may refer to a process of multiplying a transform coefficient level by a factor.

-As a result of scaling for the transform coefficient level, a transform coefficient can be generated. Scaling may also be referred to as dequantization.

Quantization Parameter (QP): The quantization parameter may mean a value used when generating a transform coefficient level for a transform coefficient in quantization. Alternatively, the quantization parameter may mean a value used when generating transform coefficients by scaling a transform coefficient level in inverse quantization. Alternatively, the quantization parameter may be a value mapped to the quantization step size.

Delta quantization parameter: The delta quantization parameter may mean a difference between a predicted quantization parameter and a quantization parameter of a target unit.

Scan: Scan may refer to a method of sorting the order of coefficients in a unit, block or matrix. For example, arranging a two-dimensional array in a one-dimensional array form may be referred to as a scan. Alternatively, arranging the one-dimensional array in the form of a two-dimensional array may also be referred to as scan or inverse scan.

Transform coefficient: The transform coefficient may be a coefficient value generated by performing transformation in the encoding device. Alternatively, the transform coefficient may be a coefficient value generated by performing at least one of entropy decoding and inverse quantization in the decoding apparatus.

-A quantized level or a quantized transform coefficient level generated by applying quantization to a transform coefficient or residual signal may also be included in the meaning of the transform coefficient.

Quantized level: The quantized level may mean a value generated by performing quantization on a transform coefficient or a residual signal in an encoding apparatus. Alternatively, the quantized level may mean a value that is an object of inverse quantization when performing inverse quantization in the decoding apparatus.

-The quantized transform coefficient level resulting from transform and quantization may also be included in the meaning of the quantized level.

Non-zero transform coefficient: The non-zero transform coefficient may mean a transform coefficient having a non-zero value or a transform coefficient level having a non-zero value. Alternatively, the non-zero transform coefficient may mean a transform coefficient whose value size is not 0 or a transform coefficient level whose value size is not 0.

Quantization matrix: A quantization matrix may mean a matrix used in a quantization process or an inverse quantization process to improve the subjective or objective picture quality of an image. The quantization matrix can also be referred to as a scaling list.

Quantization matrix coefficient: The quantization matrix coefficient may mean each element in the quantization matrix. The quantization matrix coefficient may also be referred to as a matrix coefficient.

Default matrix: The default matrix may be a quantization matrix defined in the encoding device and the decoding device.

Non-default matrix: The non-default matrix may be a quantization matrix that is not defined in the encoding apparatus and the decoding apparatus. The non-default matrix may mean a quantization matrix signaled by a user from an encoding device to a decoding device.

Most Probable Mode (MPM): MPM may indicate an intra prediction mode that is likely to be used for intra prediction of a target block.

-The encoding device and the decoding device may determine one or more MPMs based on coding parameters related to the target block and attributes of an entity related to the target block.

-The encoding device and the decoding device may determine one or more MPMs based on the intra prediction mode of the reference block. The reference block may be plural. The plurality of reference blocks may include a spatial neighboring block adjacent to the left side of the target block and a spatial neighboring block adjacent to the top of the target block. In other words, different one or more MPMs may be determined according to which intra prediction modes are used for reference blocks.

-One or more MPMs may be determined in the same manner in the encoding device and the decoding device. In other words, the encoding device and the decoding device may share an MPM list including the same one or more MPMs.

MPM list: The MPM list may be a list including one or more MPMs. The number of one or more MPMs in the MPM list may be predefined.

MPM indicator: The MPM indicator may indicate an MPM used for intra prediction of a target block among one or more MPMs in the MPM list. For example, the MPM indicator may be an index to the MPM list.

-Since the MPM list is determined in the same manner by the encoding device and the decoding device, the MPM list itself may not need to be transmitted from the encoding device to the decoding device.

-The MPM indicator may be signaled from the encoding device to the decoding device. As the MPM indicator is signaled, the decoding apparatus may determine an MPM to be used for intra prediction of a target block among MPMs in the MPM list.

MPM usage indicator: The MPM usage indicator may indicate whether the MPM usage mode is used for prediction for the target block. The MPM use mode may be a mode for determining an MPM to be used for intra prediction for a target block using an MPM list.

-The MPM usage indicator may be signaled from the encoding device to the decoding device.

Signaling: Signaling may indicate that information is transmitted from an encoding device to a decoding device. Or, signaling may mean including information in a bitstream or recording medium. Information signaled by the encoding device may be used by the decoding device.

-The encoding device may generate encoded information by encoding the signaled information. The encoded information may be transmitted from the encoding device to the decoding device. The decoding apparatus may obtain information by performing decoding on the transmitted encoded information. Here, the encoding may be entropy encoding, and the decoding may be entropy decoding.

Statistical value: Variables, coding parameters, constants, etc., can have values that can be calculated. The statistical value may be a value generated by an operation on the values of these specified objects. For example, the statistical values are average values for values such as a specified variable, a specified coding parameter, and a specified constant, a weighted average value, a weighted sum, a minimum value, a maximum value, and a mode. It can be one or more of a value, an intermediate value and an interpolation value.

1 is a block diagram showing a configuration according to an embodiment of an encoding apparatus to which the present invention is applied.

The encoding device 100 may be an encoder, a video encoding device, or a video encoding device. A video may include one or more images. The encoding apparatus 100 may sequentially encode one or more images of a video.

Referring to FIG. 1, the encoding apparatus 100 includes an inter prediction unit 110, an intra prediction unit 120, a switch 115, a subtractor 125, a transformation unit 130, a quantization unit 140, and entropy encoding It may include a unit 150, an inverse quantization unit 160, an inverse transform unit 170, an adder 175, a filter unit 180, and a reference picture buffer 190.

The encoding apparatus 100 may perform encoding on a target image using intra mode and / or inter mode. In other words, the prediction mode for the target block may be one of intra mode and inter mode.

Hereinafter, the terms "intra mode", "intra prediction mode", "in-screen mode" and "in-screen prediction mode" may be used in the same sense, and may be used interchangeably.

Hereinafter, the terms "inter mode", "inter prediction mode", "inter-screen mode" and "inter-screen prediction mode" may be used with the same meaning, and may be used interchangeably.

Hereinafter, the term “image” may refer only to a part of the image and may refer to a block. Also, the processing for “image” may indicate sequential processing for a plurality of blocks.

In addition, the encoding apparatus 100 may generate a bitstream including information encoded through encoding for a target image, and output and store the generated bitstream. The resulting bitstream can be stored on a computer readable recording medium and streamed over a wired and / or wireless transmission medium.

As the prediction mode, when the intra mode is used, the switch 115 can be switched to intra. As a prediction mode, when the inter mode is used, the switch 115 may be switched to inter.

The encoding apparatus 100 may generate a prediction block for the target block. Also, after the prediction block is generated, the encoding apparatus 100 may encode the residual block for the target block by using the residuals of the target block and the prediction block.

When the prediction mode is an intra mode, the intra prediction unit 120 may use pixels of blocks that are already encoded and / or decoded in the neighborhood of the target block as reference samples. The intra prediction unit 120 may perform spatial prediction on the target block using the reference sample, and generate prediction samples for the target block through spatial prediction. The prediction sample may mean a sample in a prediction block.

The inter prediction unit 110 may include a motion prediction unit and a motion compensation unit.

When the prediction mode is the inter mode, the motion prediction unit may search for a region that best matches the target block from the reference image during the motion prediction process, and derive a motion vector for the target block and the searched region using the searched region. can do. At this time, the motion prediction unit may use a search area as an area to be searched.

The reference image may be stored in the reference picture buffer 190, and the encoded and / or decoded reference image may be stored in the reference picture buffer 190 when encoding and / or decoding of the reference image is processed.

As the decoded picture is stored, the reference picture buffer 190 may be a decoded picture buffer (DPB).

The motion compensation unit may generate a prediction block for the target block by performing motion compensation using a motion vector. Here, the motion vector may be a 2D vector used for inter prediction. Also, the motion vector may indicate an offset between the target image and the reference image.

When the motion vector has a non-integer value, the motion prediction unit and the motion compensation unit may generate a prediction block by applying an interpolation filter to some regions in the reference image. In order to perform inter prediction or motion compensation, a method of motion prediction and motion compensation of a PU included in a CU based on a CU is skip mode, merge mode, and advanced motion vector prediction Prediction (AMVP) mode and current picture reference mode may be determined, and inter prediction or motion compensation may be performed according to each mode.

The subtractor 125 may generate a residual block that is a difference between the target block and the prediction block. The residual block may be referred to as a residual signal.

The residual signal may mean a difference between the original signal and the predicted signal. Alternatively, the residual signal may be a signal generated by transforming or quantizing the difference between the original signal and the prediction signal, or by transforming and quantizing. The residual block may be a residual signal for a block unit.

The transform unit 130 may generate transform coefficients by performing a transform on the residual block, and output the generated transform coefficients. Here, the transform coefficient may be a coefficient value generated by performing a transform on the residual block.

The conversion unit 130 may use one of a plurality of predefined conversion methods in performing the conversion.

The predefined plurality of transform methods may include Discrete Cosine Transform (DCT), Discrete Sine Transform (DST) and Karhunen-Loeve Transform (KLT) based transform, etc. have.

The transform method used for transform for the residual block may be determined according to at least one of coding parameters for a target block and / or a neighboring block. For example, the transform method may be determined based on at least one of an inter prediction mode for a PU, an intra prediction mode for a PU, a size of a TU, and a form of a TU. Alternatively, transformation information indicating a transformation method may be signaled from the encoding apparatus 100 to the decoding apparatus 200.

When the transform skip mode is applied, the transform unit 130 may omit the transform for the residual block.

By applying quantization to a transform coefficient, a quantized transform coefficient level or a quantized level can be generated. Hereinafter, in embodiments, quantized transform coefficient levels and quantized levels may also be referred to as transform coefficients.

The quantization unit 140 may generate a quantized transform coefficient level (that is, a quantized level or a quantized coefficient) by quantizing the transform coefficient according to a quantization parameter. The quantization unit 140 may output the generated quantized transform coefficient level. At this time, the quantization unit 140 may quantize the transform coefficient using a quantization matrix.

The entropy encoding unit 150 may generate a bitstream by performing entropy encoding according to a probability distribution based on values calculated by the quantization unit 140 and / or coding parameter values calculated during an encoding process. . The entropy encoding unit 150 may output the generated bitstream.

The entropy encoding unit 150 may perform entropy encoding on information about pixels of an image and information for decoding an image. For example, information for decoding an image may include a syntax element.

When entropy encoding is applied, a small number of bits may be allocated to a symbol having a high probability of occurrence, and a large number of bits may be allocated to a symbol having a low probability of occurrence. As a symbol is expressed through such allocation, the size of a bitstring for symbols to be encoded can be reduced. Accordingly, compression performance of image encoding may be improved through entropy encoding.

In addition, the entropy encoding unit 150 may perform exponential golomb, context-adaptive variable length coding (CAVLC), and context-adaptive binary arithmetic coding for entropy encoding. Coding methods such as Arithmetic Coding (CABAC) can be used. For example, the entropy encoding unit 150 may perform entropy encoding using a variable length coding (VLC) table. For example, the entropy encoding unit 150 may derive a binarization method for a target symbol. Also, the entropy encoder 150 may derive a probability model of the target symbol / bin. The entropy encoding unit 150 may perform arithmetic encoding using the derived binarization method, probability model, and context model.

In order to encode the quantized transform coefficient level, the entropy encoding unit 150 may change the coefficient of the form of a 2D block into a form of a 1D vector through a transform coefficient scanning method.

Coding parameters may be information required for encoding and / or decoding. Coding parameters may include information that is encoded in the encoding device 100 and transmitted from the encoding device 100 to a decoding device, and may include information that may be derived in an encoding or decoding process. For example, as information transmitted to the decoding device, there is a syntax element.

The coding parameter may include information derived from an encoding process or a decoding process, as well as information (or flags and indexes) encoded by the encoding device and signaled from the encoding device to a decoding device, such as syntax elements. have. Also, the coding parameter may include information required for encoding or decoding an image. For example, the size of the unit / block, the shape of the unit / block, the depth of the unit / block, the division information of the unit / block, the division structure of the unit / block, the information indicating whether the unit / block is divided into a quad tree form, Information indicating whether a unit / block is divided into a binary tree form, a binary tree form split direction (horizontal or vertical direction), a binary tree form split form (symmetric split or asymmetric split), and a unit / block form a ternary tree form Information indicating whether or not to be divided into, ternary tree type division direction (horizontal or vertical direction), ternary tree type division type (symmetric division or asymmetric division, etc.), unit / block is a multi-type tree Information indicating whether or not to be divided into forms, combination and direction of multi-type tree type segmentation (such as horizontal or vertical), and segmentation form of multi-type tree type segmentation (symmetric) Or asymmetric split), multi-type tree split tree (binary tree or ternary tree), type of prediction mode (intra prediction or inter prediction), intra prediction mode / direction, intra luma prediction mode / direction, intra chroma prediction mode / Direction, intra-split information, inter-split information, coding block split flag, prediction block split flag, transform block split flag, reference sample filtering method, reference sample filter tab, reference sample filter coefficient, prediction block filtering method, prediction block filter tab , Prediction block filter coefficient, prediction block boundary filtering method, prediction block boundary filter tap, prediction block boundary filter coefficient, inter prediction mode, motion information, motion vector, motion vector difference, reference picture index, inter prediction direction, inter prediction indicator, Prediction list utilization flag, reference picture list, reference image, POC, motion vector predictor, motion Data prediction index, motion vector prediction candidate, motion vector candidate list, information indicating whether to use merge mode, information indicating whether to use merge index, merge candidate, merge candidate list, skip mode, interpolation filter Type, filter tab of interpolation filter, filter coefficient of interpolation filter, motion vector size, motion vector representation accuracy, transform type, transform size, information indicating whether to use the first order transform, whether to use the additional (secondary) transform Information indicating, primary transform selection information (or primary transform index), secondary transform selection information (or secondary transform index), information indicating the presence or absence of a residual signal, coded block pattern, Coded block flags, quantization parameters, residual quantization parameters, quantization matrices, information about intra-loop filters, applying intra-loop filters Information indicating whether or not to apply, intra-loop filter coefficient, intra-loop filter tap, intra loop filter shape / form, whether to apply deblocking filter, coefficient of deblocking filter , Filter tab of deblocking filter, strength of deblocking filter, shape / shape of deblocking filter, information indicating whether adaptive sample offset is applied, adaptive sample offset value, adaptive sample offset category, adaptive sample offset Type, information indicating whether to apply the adaptive in-loop filter, the coefficient of the adaptive loop-in-filter, the filter tap of the adaptive loop-in-filter, the shape / shape of the adaptive loop-in-filter , Binarization / De-binarization method, context model, context model determination method, context model update method, information indicating whether to perform the regular mode, indicating whether to perform the bypass mode Information, Significant Count Flag, Last Significant Count Flag, Count Group Unit Coding Flag, Last Significant Count Position, Flag indicating whether count value is greater than 1, Flag indicating whether count value is greater than 2, Count Flag indicating whether the value is greater than 3, rest coefficient value information, sign information, restored luma sample, restored chroma sample, context bin, bypass bin, residual luma sample, residual chroma sample, transform coefficient, Luma transform coefficient, chroma transform coefficient, quantized level, luma quantized level, chroma quantized level, transform coefficient level, luma transform coefficient level, chroma transform coefficient level, transform coefficient level scanning method, motion vector on the side of the decoding apparatus Size of search area, motion on the side of the decoding device Vector type of search area, motion on the side of the decoding device Vector search count, CTU size, minimum block size, maximum block size, maximum block depth, minimum block depth, display / output order of images, slice identification information, slice type, slice segmentation information, tile group identification information, tile group type, Tile group segmentation information, tile identification information, tile type, tile segmentation information, picture type, bit depth, input sample bit depth, reconstructed sample bit depth, residual sample bit depth, transform coefficient bit depth, quantized level bit depth, Coding parameters may include at least one value, a combined form, or statistics of information about a luma signal and information about a chroma signal. Also, information related to the above-described coding parameters may be included in the coding parameters. Information used to calculate and / or derive the above-described coding parameters may also be included in the coding parameters. Information calculated or derived using the above-described coding parameters may also be included in the coding parameters.

The prediction scheme may indicate one prediction mode of intra prediction mode and inter prediction mode.

The primary transform selection information may indicate a primary transform applied to the target block.

The secondary transform selection information may indicate a secondary transform applied to the target block.

The residual signal may represent a difference between the original signal and the prediction signal. Alternatively, the residual signal may be a signal generated by transforming a difference between the original signal and the prediction signal. Alternatively, the residual signal may be a signal generated by transforming and quantizing the difference between the original signal and the prediction signal. The residual block may be a residual signal for the block.

Here, signaling information may mean that the encoding apparatus 100 includes entropy-encoded information generated by performing entropy encoding on a flag or index in a bitstream, The decoding apparatus 200 may obtain information by performing entropy decoding on the entropy-encoded information extracted from the bitstream. Here, the information may include flags and indexes.

Since encoding through inter prediction is performed by the encoding apparatus 100, the encoded target image may be used as a reference image for other image (s) to be processed later. Accordingly, the encoding apparatus 100 may reconstruct or decode the encoded target image again, and may store the reconstructed or decoded image in the reference picture buffer 190 as a reference image. Inverse quantization and inverse transform on the encoded target image for decoding may be processed.

The quantized level may be inversely quantized in the inverse quantization unit 160 and inversely transformed in the inverse transformation unit 170. The inverse quantization unit 160 may generate an inverse quantized coefficient by performing inverse quantization on a quantized level. The inverse transform unit 170 may generate an inverse quantized and inverse transformed coefficient by performing an inverse transform on the inverse quantized coefficient.

The inverse quantized and inverse transformed coefficients may be combined with the prediction block through the adder 175, and the reconstructed block may be generated by adding the inverse quantized and inverse transformed coefficients and the prediction block. Here, the inverse quantized and / or inverse transformed coefficient may mean a coefficient in which at least one of dequantization and inverse-transformation has been performed, and may mean a reconstructed residual block.

The reconstructed block may pass through the filter unit 180. The filter unit 180 includes at least one of a deblocking filter, a sample adaptive offset (SAO), an adaptive loop filter (ALF), and a non-local filter (NLF). One or more may be applied to a reconstructed sample, reconstructed block, or reconstructed picture. The filter unit 180 may also be referred to as an in-loop filter.

The deblocking filter can remove block distortion occurring at the boundary between blocks. In order to determine whether to apply the deblocking filter, it may be determined whether to apply the deblocking filter to the target block based on the pixel (s) included in several columns or rows included in the block.

When a deblocking filter is applied to a target block, the applied filter may vary depending on the strength of deblocking filtering required. That is, a filter determined according to the intensity of deblocking filtering among different filters may be applied to the target block. When a deblocking filter is applied to a target block, one of a strong filter and a weak filter may be applied to the target block according to the required strength of deblocking filtering.

Also, when vertical filtering and horizontal filtering are performed on the target block, horizontal filtering and vertical filtering may be processed in parallel.

The SAO can add an appropriate offset to the pixel value of the pixel to compensate for coding errors. The SAO may perform correction using an offset on a difference between an original image and a deblocking-applied image in units of pixels with respect to the deblocking-applied image. In order to perform offset correction for an image, a method in which pixels included in the image are divided into a certain number of regions, and then an area to be offset among the divided regions is determined and an offset is applied to the determined region can be used. In addition, a method of applying an offset in consideration of edge information of each pixel of an image may be used.

The ALF can perform filtering based on a comparison value between the reconstructed image and the original image. After dividing the pixels included in the image into predetermined groups, a filter to be applied to each divided group may be determined, and filtering may be differentially performed for each group. Information related to whether to apply the adaptive loop filter may be signaled for each CU. This information can be signaled for the luma signal. The shape and filter coefficient of ALF to be applied to each block may be different for each block. Alternatively, regardless of the characteristics of the block, a fixed form of ALF can be applied to the block.

The non-local filter can perform filtering based on reconstructed blocks similar to the target block. In the reconstructed image, an area similar to the target block may be selected, and filtering of the target block may be performed using statistical properties of the selected similar area. Information related to whether to apply a non-local filter may be signaled for the CU. Also, the shapes and filter coefficients of the non-local filter to be applied to the blocks may be different depending on the block.

The reconstructed block or the reconstructed image that has passed through the filter unit 180 may be stored in the reference picture buffer 190 as a reference picture. The reconstructed block that has passed through the filter unit 180 may be part of a reference picture. In other words, the reference picture may be a reconstructed picture composed of reconstructed blocks that have passed through the filter unit 180. The stored reference picture can then be used for inter prediction or motion compensation.

2 is a block diagram showing a configuration according to an embodiment of a decoding apparatus to which the present invention is applied.

The decoding apparatus 200 may be a decoder, a video decoding apparatus, or a video decoding apparatus.

Referring to FIG. 2, the decoding apparatus 200 includes an entropy decoding unit 210, an inverse quantization unit 220, an inverse transform unit 230, an intra prediction unit 240, an inter prediction unit 250, and a switch 245. , An adder 255, a filter unit 260 and a reference picture buffer 270.

The decoding apparatus 200 may receive a bitstream output from the encoding apparatus 100. The decoding apparatus 200 may receive a bitstream stored in a computer-readable recording medium, and may receive a bitstream streamed through a wired / wireless transmission medium.

The decoding apparatus 200 may perform intra-mode and / or inter-mode decoding on the bitstream. Also, the decoding apparatus 200 may generate a reconstructed image or a decoded image through decoding, and may output the generated reconstructed image or decoded image.

For example, switching to the intra mode or the inter mode according to the prediction mode used for decoding may be performed by the switch 245. When the prediction mode used for decoding is the intra mode, the switch 245 may be switched to the intra mode. When the prediction mode used for decoding is an inter mode, the switch 245 may be switched to inter.

The decoding apparatus 200 may obtain a reconstructed residual block by decoding the input bitstream, and generate a prediction block. When the reconstructed residual block and the prediction block are obtained, the decoding apparatus 200 may generate a reconstructed block to be decoded by adding the reconstructed residual block and the prediction block.

The entropy decoding unit 210 may generate symbols by performing entropy decoding on the bitstream based on the probability distribution for the bitstream. The generated symbols may include a symbol in the form of a quantized transform coefficient level (that is, a quantized level or a quantized coefficient). Here, the entropy decoding method may be similar to the entropy encoding method described above. For example, the entropy decoding method may be an inverse process of the entropy encoding method described above.

In order to decode the quantized transform coefficient level, the entropy decoding unit 210 may change the coefficient in the form of a one-dimensional vector into the form of a two-dimensional block through a transform coefficient scanning method.

For example, the coefficients may be changed into a two-dimensional block form by scanning the coefficients of the block using the upper right diagonal scan. Or, depending on the size of the block and / or the intra prediction mode, which of the upper right diagonal scan, vertical scan and horizontal scan may be used.

The quantized coefficients may be inverse quantized by the inverse quantization unit 220. The inverse quantization unit 220 may generate an inverse quantized coefficient by performing inverse quantization on the quantized coefficient. Also, the inverse quantized coefficient may be inverse transformed by the inverse transform unit 230. The inverse transform unit 230 may generate a reconstructed residual block by performing inverse transform on an inverse quantized coefficient. As a result of inverse quantization and inverse transformation on the quantized coefficients, a reconstructed residual block may be generated. At this time, the inverse quantization unit 220 may apply a quantization matrix to the quantized coefficients in generating the reconstructed residual block.

When the intra mode is used, the intra prediction unit 240 may generate a prediction block by performing spatial prediction on the target block using pixel values of the already decoded block of the neighbor of the target block.

The inter prediction unit 250 may include a motion compensation unit. Alternatively, the inter prediction unit 250 may be referred to as a motion compensation unit.

When the inter mode is used, the motion compensation unit may generate a prediction block by performing motion compensation on a target block using a motion vector and a reference image stored in the reference picture buffer 270.

When the motion vector has a non-integer value, the motion compensation unit may apply an interpolation filter to some areas in the reference image, and generate a prediction block using the reference image to which the interpolation filter is applied. The motion compensation unit may determine which of the skip mode, the merge mode, the AMVP mode, and the current picture reference mode is a motion compensation method used for the PU included in the CU based on the CU to perform motion compensation. Depending on the motion compensation can be performed.

The reconstructed residual block and prediction block may be added through the adder 255. The adder 255 may generate a reconstructed block by adding a reconstructed residual block and a prediction block.

The reconstructed block may pass through the filter unit 260. The filter unit 260 may apply at least one of a deblocking filter, a SAO, an ALF, and a non-local filter to a reconstructed block or a reconstructed image. The reconstructed image may be a picture including a reconstructed block.

The filter unit 260 may output the reconstructed image.

The reconstructed block and / or reconstructed image that has passed through the filter unit 260 may be stored as a reference picture in the reference picture buffer 270. The reconstructed block that has passed through the filter unit 260 may be part of the reference picture. In other words, the reference picture may be a reconstructed image composed of reconstructed blocks that have passed through the filter unit 260. The stored reference picture can then be used for inter prediction and / or motion compensation.

FIG. 3 is a diagram schematically showing an image segmentation structure when encoding and decoding an image.

3 may schematically show an example in which one unit is divided into a plurality of sub-units.

In order to efficiently divide an image, in coding and decoding, a coding unit (CU) may be used. The unit may be a term referring to 1) a block including image samples and 2) a syntax element. For example, "division of a unit" may mean "division of a block corresponding to a unit".

A CU may be used as a base unit for image encoding and / or decoding. Further, the CU may be used as a unit to which one selected mode of intra mode and inter mode is applied in image encoding and / or decoding. That is, in video encoding and / or decoding, it may be determined which of the intra mode and the inter mode is applied to each CU.

Also, the CU may be a base unit in encoding and / or decoding prediction, transform, quantization, inverse transform, inverse quantization, and transform coefficients.

Referring to FIG. 3, the image 300 may be sequentially divided into units of a largest coding unit (LCU). For each LCU, a split structure can be determined. Here, LCU may be used in the same sense as Coding Tree Unit (CTU).

The division of the unit may mean division of a block corresponding to the unit. The block segmentation information may include depth information regarding the depth of the unit. The depth information may indicate the number and / or degree of division of the unit. One unit may be divided into a plurality of sub-units hierarchically with depth information based on a tree structure.

Each divided sub-unit may have depth information. The depth information may be information indicating the size of the CU. Depth information may be stored for each CU.

Each CU may have depth information. When a CU is divided, CUs generated by partitioning may have an increased depth by 1 from the depth of the divided CU.

The split structure may mean a distribution of a CU in the LCU 310 for efficiently encoding an image. This distribution may be determined according to whether to divide one CU into a plurality of CUs. The number of divided CUs may be two or more positive integers including 2, 4, 8, and 16.

The horizontal and vertical sizes of the CU generated by partitioning may be smaller than the horizontal and vertical sizes of the CU before partitioning, depending on the number of CUs generated by partitioning. For example, the horizontal size and vertical size of a CU generated by division may be half the horizontal size and half the vertical size of the CU before division.

The divided CU can be recursively divided into a plurality of CUs in the same way. By recursive partitioning, a size of at least one of a horizontal size and a vertical size of a divided CU may be reduced compared to at least one of a horizontal size and a vertical size of a CU before partitioning.

The division of the CU can be made recursively up to a predefined depth or a predefined size.

For example, the depth of the CU may have a value of 0 to 3. The size of the CU may be from 64x64 to 8x8 depending on the depth of the CU.

For example, the depth of the LCU 310 may be 0, and the depth of the smallest coding unit (SCU) may be a predefined maximum depth. Here, the LCU may be a CU having the largest coding unit size as described above, and the SCU may be a CU having the smallest coding unit size.

Splitting may be started from the LCU 310, and the depth of the CU may increase by 1 whenever the horizontal and / or vertical size of the CU is reduced by the splitting.

For example, for each depth, a non-divided CU may have a size of 2Nx2N. Also, in the case of a divided CU, a 2Nx2N sized CU may be divided into 4 CUs having an NxN size. The size of N can be reduced in half with each increase of depth.

Referring to FIG. 3, an LCU having a depth of 0 may be 64 × 64 pixels or 64 × 64 blocks. 0 can be the minimum depth. The SCU with a depth of 3 may be 8x8 pixels or 8x8 blocks. 3 can be the maximum depth. At this time, the CU of the 64x64 block, which is an LCU, may be represented by a depth of 0. A CU of a 32x32 block can be represented by depth 1. The CU of a 16x16 block may be represented by depth 2. The CU of the 8x8 block, which is the SCU, may be represented by depth 3.

Information on whether the CU is divided may be expressed through partitioning information of the CU. The segmentation information may be 1-bit information. All CUs except the SCU may include segmentation information. For example, a value of split information of a CU that is not split may be a first value, and a value of split information of a CU that is split may be a second value. When the split information indicates whether the CU is split, the first value may be 0, and the second value may be 1.

For example, when one CU is divided into four CUs, the horizontal size and vertical size of each CU of the four CUs generated by the partition are half the horizontal size and the half the vertical size of the CU before division, respectively. Can be. When a 32x32 sized CU is divided into 4 CUs, the size of the divided 4 CUs may be 16x16. When one CU is divided into four CUs, it can be said that the CU is divided into quad-trees. In other words, quad-tree partition is applied to the CU.

For example, when one CU is divided into two CUs, the horizontal size or vertical size of each CU of the two CUs generated by the partition is half the horizontal size or half the vertical size of the CU before partitioning, respectively. Can be. When a CU of 32x32 size is vertically divided into two CUs, the size of the divided two CUs may be 16x32. When a CU of 32x32 size is horizontally divided into two CUs, the sizes of the divided two CUs may be 32x16. When one CU is divided into two CUs, it can be said that the CU is divided into a binary-tree type. In other words, binary-tree partition is applied to CU.

For example, when one CU is divided into three CUs, three divided CUs may be generated by dividing the horizontal or vertical size of the CU before division by a ratio of 1: 2: 1. For example, when a 16x32 sized CU is divided into three CUs in the horizontal direction, the divided three CUs may have a size of 16x8, 16x16, and 16x8, respectively, from the top. For example, when a 32x32 sized CU is divided into three CUs in the vertical direction, the divided three CUs may have a size of 8x32, 16x32, and 8x32, respectively, from the left. When one CU is divided into three CUs, it can be said that the CU is divided into a ternary-tree form. In other words, it can be seen that a ternary-tree partition was applied to the CU.

In the LCU 310 of FIG. 3, both quad-tree type division and binary-tree type division are applied.

In the encoding apparatus 100, a coding tree unit (CTU) having a size of 64x64 may be divided into a plurality of smaller CUs by a recursive quad-tree structure. One CU can be divided into 4 CUs having the same size. CUs may be recursively partitioned, and each CU may have a quad tree structure.

Through recursive partitioning for the CU, an optimal partitioning method that generates the minimum rate-distortion ratio can be selected.

The CTU 320 of FIG. 3 is an example of a CTU in which quad tree splitting, binary tree splitting, and ternary tree splitting are all applied.

As described above, in order to divide the CTU, at least one of quad tree split, binary tree split and ternary tree split may be applied to the CTU. Splits can be applied based on a specified priority.

For example, quad tree splitting may be preferentially applied to CTU. A CU that can no longer be split into a quad tree may correspond to a leaf node of the quad tree. The CU corresponding to the leaf node of the quad tree may be a binary tree and / or a root node of the ternary tree. That is, the CU corresponding to the leaf node of the quad tree may be divided into a binary tree shape or a ternary tree shape, or may not be further divided. At this time, the quad tree split is not applied to the CU generated by applying binary tree splitting or ternary tree splitting to the CU corresponding to the leaf node of the quadtree, so that the block splitting and / or signaling of block splitting information is prevented. It can be done effectively.

The division of the CU corresponding to each node of the quad tree can be signaled using the quad division information. The quad split information having a first value (eg, '1') may indicate that the CU is split in a quad tree format. The quad splitting information having a second value (eg, '0') may indicate that the CU is not split in a quad tree format. The quad segmentation information may be a flag having a specified length (eg, 1 bit).

Priority may not exist between binary tree division and ternary tree division. That is, the CU corresponding to the leaf node of the quad tree may be divided into a binary tree form or a ternary tree form. Also, a CU generated by binary tree division or ternary tree division may be divided into a binary tree shape or a ternary tree shape, or may not be further divided.

Partitioning when there is no priority between binary tree partitioning and ternary tree partitioning may be referred to as multi-type tree partitioning. That is, a CU corresponding to a leaf node of a quad tree may be a root node of a multi-type tree. For the partitioning of a CU corresponding to each node of a multi-type tree, it may be signaled using at least one of information indicating whether a multi-type tree is split, split direction information, and split tree information. In order to divide a CU corresponding to each node of a multi-type tree, information indicating whether to sequentially divide, split direction information, and split tree information may be signaled.

For example, information indicating whether to divide a multi-type tree having a first value (eg, “1”) may indicate that the CU is divided into a multi-type tree. Information indicating whether to divide the multi-type tree having the second value (eg, “0”) may indicate that the corresponding CU is not divided into a multi-type tree.

When the CU corresponding to each node of the multi-type tree is divided into a multi-type tree, the CU may further include split direction information.

The split direction information may indicate a split direction of multi-type tree splitting. The split direction information having the first value (eg, “1”) may indicate that the corresponding CU is split in the vertical direction. The split direction information having the second value (eg, “0”) may indicate that the CU is split in the horizontal direction.

When the CU corresponding to each node of the multi-type tree is divided into a multi-type tree, the CU may further include split tree information. The split tree information may indicate a tree used for multi-type tree splitting.

For example, split tree information having a first value (eg, “1”) may indicate that the CU is split into a binary tree. The split tree information having the second value (eg, “0”) may indicate that the CU is split into a ternary tree shape.

Here, each of the information indicating whether or not the above-described partitioning, partitioning tree information, and partitioning direction information may be flags having a specified length (for example, 1 bit).

At least one of the above-described quad splitting information, information indicating whether to split the multi-type tree, splitting direction information, and splitting tree information may be entropy encoded and / or entropy decoded. For entropy encoding / decoding of such information, information of a neighboring CU adjacent to the target CU may be used.

For example, the split form of the left CU and / or the upper CU (that is, whether to split, split tree and / or split direction) and the split form of the target CU may be considered to have a high probability of being similar to each other. Accordingly, context information for entropy encoding and / or entropy decoding of information of the target CU may be derived based on the information of the neighboring CU. At this time, the information of the neighboring CU may include at least one of 1) quad splitting information of the neighboring CU, 2) information indicating whether to divide the multi-type tree, 3) splitting direction information, and 4) splitting tree information.

As another embodiment, among binary tree division and ternary tree division, binary tree division may be preferentially performed. That is, binary tree splitting is applied first, and a CU corresponding to a leaf node of the binary tree may be set as the root node of the ternary tree. In this case, quad tree splitting and binary tree splitting may not be performed for the CU corresponding to the node of the ternary tree.

A CU that is no longer split by quad tree splitting, binary tree splitting, and / or ternary tree splitting may be a unit of coding, prediction, and / or transformation. That is, for prediction and / or transformation, the CU may no longer be split. Therefore, a segmentation structure, segmentation information, etc. for dividing a CU into a prediction unit and / or a transformation unit may not exist in the bitstream.

However, if the size of the CU that is a unit of division is larger than the size of the maximum transform block, these CUs may be recursively partitioned until the size of the CU is less than or equal to the size of the maximum transform block. For example, when the size of the CU is 64x64 and the size of the maximum transform block is 32x32, the CU may be divided into four 32x32 blocks for conversion. For example, if the size of the CU is 32x64 and the size of the maximum transform block is 32x32, the CU may be divided into two 32x32 blocks for conversion.

In this case, information on whether the CU is split for conversion may not be signaled separately. Without signaling, whether to divide the CU may be determined by comparison between the horizontal size (and / or vertical size) of the CU and the horizontal size (and / or vertical size) of the largest transform block. For example, if the horizontal size of the CU is larger than the horizontal size of the maximum transform block, the CU may be divided into two equal parts vertically. In addition, if the vertical size of the CU is larger than the vertical size of the maximum transform block, the CU may be divided into two horizontally.

Information regarding the maximum size and / or minimum size of the CU, and information about the maximum size and / or minimum size of the transform block may be signaled or determined at a higher level for the CU. For example, the higher level may be a sequence level, a picture level, a tile level, a tile group level and a slice level. For example, the minimum size of a CU may be determined as 4x4. For example, the maximum size of the transform block may be determined as 64x64. For example, the minimum size of the transform block may be determined as 4x4.

Information about the minimum size of the CU corresponding to the leaf node of the quad tree (that is, the minimum size of the quad tree) and / or the maximum depth of the path from the root node to the leaf node of the multi-type tree (that is, the multi-type tree maximum) Depth) information may be signaled or determined at a higher level for the CU. For example, the higher level may be a sequence level, a picture level, a slice level, a tile group level and a tile level. The information about the quad tree minimum size and / or the information about the multi-type tree maximum depth may be signaled or determined separately for each of the intra intra slice and the inter slice.

The difference information about the size of the CTU and the maximum size of the transform block may be signaled or determined at a higher level for the CU. For example, the higher level may be a sequence level, a picture level, a slice level, a tile group level and a tile level. Information about the maximum size of the CU corresponding to each node of the binary tree (that is, the maximum size of the binary tree) may be determined based on the size and difference information of the CTU. The maximum size of the CU corresponding to each node of the ternary tree (that is, the maximum size of the ternary tree) may have a different value depending on the type of slice. For example, within an intra slice, the maximum size of a striking tree may be 32x32. Also, for example, within an inter slice, the maximum size of a ternary tree may be 128x128. For example, the minimum size of the CU corresponding to each node of the binary tree (that is, the minimum size of the binary tree) and / or the minimum size of the CU corresponding to each node of the ternary tree (that is, the minimum size of the ternary tree) is the CU It can be set to the minimum size.

As another example, the binary tree maximum size and / or the ternary tree maximum size may be signaled or determined at the slice level. Also, the binary tree minimum size and / or the ternary tree minimum size may be signaled or determined at the slice level.

Based on the various block sizes and various depths described above, quad splitting information, information indicating whether to split the multi-type tree, split tree information, and / or split direction information may or may not be present in the bitstream.

For example, if the size of the CU is not larger than the minimum size of the quad tree, the CU may not include quad splitting information, and quad splitting information for the CU may be inferred as a second value.

For example, the size (horizontal size and vertical size) of a CU corresponding to a node of a multi-type tree is greater than the maximum size of the binary tree (horizontal size and vertical size) and / or the maximum size of the ternary tree (horizontal size and vertical size). In the larger case, the CU may not be divided into binary tree form and / or ternary tree form. According to this determination method, information indicating whether to divide the multi-type tree may not be signaled and may be inferred as a second value.

Or, the size (horizontal size and vertical size) of the CU corresponding to the nodes of the multi-type tree is the same as the minimum size (horizontal size and vertical size) of the binary tree, or the size of the CU (horizontal size and vertical size) is struck out. If it is equal to twice the minimum size (horizontal size and vertical size), the CU may not be divided into binary tree form and / or ternary tree form. According to this determination method, information indicating whether to divide the multi-type tree may not be signaled and may be inferred as a second value. This is because when a CU is divided into a binary tree form and / or a ternary tree form, a CU smaller than the minimum size of the binary tree and / or the minimum size of the ternary tree is generated.

Alternatively, binary tree partitioning or ternary tree partitioning may be limited based on the size of the virtual pipeline data unit (ie, the pipeline buffer size). For example, when a CU is divided into sub-CUs that are not suitable for a pipeline buffer size, by binary tree division or ternary tree division, binary tree division or ternary tree division may be limited. The pipeline buffer size may be the same as the size of the largest transform block (eg 64X64).

For example, when the pipeline buffer size is 64X64, the following partitions may be limited.

-Splitting a ternary tree for NxM (N and / or M is 128) CU

-Horizontal binary tree splitting for 128xN (N <= 64) CU

-Vertical Binary Tree Split for Nx128 (N <= 64) CU

Alternatively, if the depth in the multi-type tree of the CU corresponding to the node of the multi-type tree is equal to the maximum depth of the multi-type tree, the CU may not be divided into a binary tree shape and / or a ternary tree shape. According to this determination method, information indicating whether to divide the multi-type tree may not be signaled and may be inferred as a second value.

Or, for a CU corresponding to a node of a multi-type tree, a multi-type tree only if at least one of a vertical binary tree split, a horizontal binary tree split, a vertical ternary chop tree split, and a horizontal ternary tree split is possible. Information indicating whether or not to be split may be signaled. Otherwise, the CU may not be split into binary tree form and / or ternary tree form. According to this determination method, information indicating whether to divide the multi-type tree may not be signaled and may be inferred as a second value.

Or, for the CU corresponding to a node of a multi-type tree, the split direction information is only possible when both vertical binary tree splitting and horizontal binary tree splitting are possible, or both vertical ternary tree splitting and horizontal ternary tree splitting are possible. Can be signaled. Otherwise, the split direction information may not be signaled and may be inferred as a value indicating the direction in which the CU can be split.

Or, for a CU corresponding to a node of a multi-type tree, split tree information is performed only when both vertical binary tree splitting and vertical ternary tree splitting are possible, or both horizontal binary tree splitting and horizontal ternary tree splitting are possible. Can be signaled. Otherwise, the split tree information may not be signaled and may be inferred as a value indicating a tree that can be applied to splitting of the CU.

FIG. 4 is a diagram illustrating a form of a prediction unit PU that the coding unit CU may include.

A CU that is no longer split among CUs split from the LCU may be divided into one or more prediction units (PUs).

The PU may be a basic unit for prediction. The PU may be encoded and decoded in any one of skip mode, inter mode, and intra mode. The PU may be divided into various forms according to each mode. For example, the object block described above with reference to FIG. 1 and the object block described with reference to FIG. 2 may be a PU.

CU may not be divided into PUs. When the CU is not divided into PUs, the size of the CU and the size of the PU may be the same.

In the skip mode, there may not be a partition in the CU. In the skip mode, a 2Nx2N mode 410 having the same size of the PU and CU without splitting may be supported.

In the inter mode, 8 types of partitions in a CU can be supported. For example, in inter mode, 2Nx2N mode 410, 2NxN mode 415, Nx2N mode 420, NxN mode 425, 2NxnU mode 430, 2NxnD mode 435, nLx2N mode 440, and nRx2N Mode 445 may be supported.

In the intra mode, 2Nx2N mode 410 and NxN mode 425 may be supported.

In 2Nx2N mode 410, a PU having a size of 2Nx2N may be encoded. A PU having a size of 2Nx2N may mean a PU having the same size as a CU. For example, a PU having a size of 2Nx2N may have a size of 64x64, 32x32, 16x16 or 8x8.

In the NxN mode 425, a PU having a size of NxN may be encoded.

For example, in intra prediction, when the size of a PU is 8x8, four divided PUs can be coded. The size of the divided PU may be 4x4.

When the PU is coded by the intra mode, the PU may be coded using one intra prediction mode among a plurality of intra prediction modes. For example, in High Efficiency Video Coding (HEVC) technology, 35 intra prediction modes may be provided, and a PU may be encoded in one of 35 intra prediction modes.

Which mode of the PU to be encoded by the 2Nx2N mode 410 or the NxN mode 425 may be determined by a rate-distortion cost.

The encoding apparatus 100 may perform encoding operations on a PU having a size of 2Nx2N. Here, the encoding operation may be to encode a PU in each of a plurality of intra prediction modes that the encoding apparatus 100 can use. An optimal intra prediction mode for a PU having a size of 2Nx2N may be derived through encoding operations. The optimal intra prediction mode may be an intra prediction mode that generates a minimum rate-distortion cost for encoding of a PU having a size of 2Nx2N among a plurality of intra prediction modes that the encoding apparatus 100 can use.

Also, the encoding apparatus 100 may sequentially perform encoding operations on each PU of PUs divided by NxN. Here, the encoding operation may be to encode a PU in each of a plurality of intra prediction modes that the encoding apparatus 100 can use. The optimal intra prediction mode for the NxN-sized PU can be derived through the encoding operation. The optimal intra prediction mode may be an intra prediction mode in which a minimum rate-distortion cost is generated for encoding of an NxN-size PU among a plurality of intra prediction modes that the encoding apparatus 100 can use.

The encoding apparatus 100 may determine which of the 2Nx2N size PU and the NxN size PU to be encoded based on the comparison of the rate-distortion cost of the 2Nx2N size PU and the rate-distortion costs of the NxN size PUs.

One CU may be divided into one or more PUs, and the PU may also be divided into a plurality of PUs.

For example, when one PU is divided into four PUs, the horizontal size and vertical size of each PU of the four PUs generated by the division are half of the horizontal size of the PU and half of the vertical size, respectively, before division. Can be. When a 32x32 sized PU is divided into 4 PUs, the size of the divided 4 PUs may be 16x16. When one PU is divided into four PUs, it can be said that the PU is divided into quad-trees.

For example, when one PU is divided into two PUs, the horizontal size or vertical size of each PU of the two PUs generated by the division is half the horizontal size or half the vertical size of the PU before division, respectively. Can be. When a 32x32 sized PU is vertically divided into two PUs, the size of the divided two PUs may be 16x32. When a 32x32 sized PU is horizontally divided into two PUs, the size of the divided two PUs may be 32x16. When one PU is divided into two PUs, it can be said that the PU is divided into a binary-tree form.

FIG. 5 is a diagram illustrating a form of a transform unit (TU) that may be included in a coding unit (CU).

A transform unit (TU) may be a basic unit used in the process of transform, quantization, inverse transform, inverse quantization, entropy encoding, and entropy decoding in a CU.

The TU may have a square shape or a rectangular shape. The shape of the TU may be determined depending on the size and / or shape of the CU.

Among CUs divided from an LCU, a CU that is no longer divided into CUs may be divided into one or more TUs. At this time, the split structure of the TU may be a quad-tree structure. For example, as illustrated in FIG. 5, one CU 510 may be divided once or more according to a quad-tree structure. Through partitioning, one CU 510 may be composed of TUs of various sizes.

If one CU is split more than once, the CU can be considered to be split recursively. Through partitioning, one CU may be composed of TUs having various sizes.

Alternatively, one CU may be divided into one or more TUs based on the number of vertical and / or horizontal lines dividing the CU.

The CU may be divided into symmetrical TUs, or may be divided into asymmetrical TUs. For division into asymmetric TUs, information on the size and / or shape of the TU may be signaled from the encoding device 100 to the decoding device 200. Alternatively, the size and / or shape of the TU may be derived from information about the size and / or shape of the CU.

CU may not be divided into TUs. If the CU is not divided into TUs, the size of the CU and the size of the TU may be the same.

One CU may be divided into one or more TUs, and the TU may also be divided into a plurality of TUs.

For example, when one TU is divided into four TUs, the horizontal size and vertical size of each TU of the four TUs generated by the division are each half the horizontal size and half the vertical size of the TU before division. Can be. When a 32x32 sized TU is divided into 4 TUs, the size of the divided 4 TUs may be 16x16. When one TU is divided into four TUs, it can be said that the TU is divided into quad-trees.

For example, when one TU is divided into two TUs, the horizontal size or vertical size of each TU of the two TUs generated by the division is half the horizontal size or half the vertical size of the TU before each split. Can be. When a 32x32 sized TU is vertically divided into two TUs, the size of the divided two TUs may be 16x32. When a 32x32 sized TU is horizontally divided into two TUs, the size of the divided two TUs may be 32x16. When one TU is divided into two TUs, it can be said that the TU is divided into a binary-tree form.

The CU may be divided in a manner other than that shown in FIG. 5.

For example, one CU may be divided into three CUs. The horizontal or vertical size of the divided three CUs may be 1/4, 1/2, and 1/4 of the horizontal or vertical size of the CU before division, respectively.

For example, when a 32x32 CU is vertically divided into 3 CUs, the size of the divided 3 CUs may be 8x32, 16x32, and 8x32, respectively. As described above, when one CU is divided into three CUs, it can be seen that the CU is divided into a ternary tree.

One of the illustrated quad tree type partitioning, binary tree type partitioning, and ternary tree type partitioning may be applied for CU partitioning, and a plurality of partitioning schemes may be combined together to be used for CU partitioning. . In this case, a case in which a plurality of splitting methods are used in combination may be referred to as splitting in the form of a composite tree.

6 illustrates division of a block according to an example.

In the process of encoding and / or decoding an image, a target block may be divided as shown in FIG. 6. For example, the target block may be a CU.

In order to divide the target block, an indicator indicating split information may be signaled from the encoding apparatus 100 to the decoding apparatus 200. The segmentation information may be information indicating how the target block is divided.

Split information includes a split flag (hereinafter referred to as "split_flag"), a quad-binary flag (hereinafter referred to as "QB_flag"), a quad tree flag (hereinafter referred to as "quadtree_flag"), and a binary tree flag (hereinafter referred to as "binarytree_flag") It may be one or more of a binary type flag (hereinafter referred to as "Btype_flag").

split_flag may be a flag indicating whether a block is split. For example, a value of 1 of split_flag may indicate that a block is split. A value of 0 of split_flag may indicate that a block is not split.

QB_flag may be a flag indicating whether a block is divided into a quad tree shape and a binary tree shape. For example, the value 0 of QB_flag may indicate that the block is divided into quad trees. The value 1 of QB_flag may indicate that the block is divided into a binary tree. Alternatively, the value 0 of QB_flag may indicate that the block is divided into binary tree shapes. The value 1 of QB_flag may indicate that the block is divided into quad trees.

quadtree_flag may be a flag indicating whether a block is divided into quad trees. For example, a value of 1 of quadtree_flag may indicate that a block is divided into quad trees. A value of 0 of quadtree_flag may indicate that a block is not divided into quad tree types.

binarytree_flag may be a flag indicating whether a block is divided into a binary tree. For example, a value of 1 of binarytree_flag may indicate that a block is divided into binary tree shapes. A value of 0 of binarytree_flag may indicate that a block is not split into binary tree form.

Btype_flag may be a flag indicating whether the block is divided into a vertical split or a horizontal split when the block is split into a binary tree. For example, the value 0 of Btype_flag may indicate that the block is divided in the horizontal direction. The value 1 of Btype_flag may indicate that the block is divided in the vertical direction. Alternatively, the value 0 of Btype_flag may indicate that the block is divided in the vertical direction. The value 1 of Btype_flag may indicate that the block is divided in the horizontal direction.

For example, the segmentation information for the block of FIG. 6 can be derived by signaling at least one of quadtree_flag, binarytree_flag, and Btype_flag as shown in Table 1 below.

[Table 1]

For example, split information for the block of FIG. 6 can be derived by signaling at least one of split_flag, QB_flag, and Btype_flag as shown in Table 2 below.

[Table 2]

Depending on the size and / or shape of the block, the partitioning method may be limited to a quad tree, or only a binary tree. When this restriction is applied, split_flag may be a flag indicating whether to split into a quad tree or a flag indicating whether to split into a binary tree. The size and shape of the block may be derived according to the depth information of the block, and the depth information may be signaled from the encoding device 100 to the decoding device 200.

When the size of a block falls within a specified range, only quad tree-shaped partitioning may be possible. For example, the specified range may be defined by at least one of a maximum block size and a minimum block size, which are only possible in quad tree-type partitioning.

Information indicating the maximum block size and / or the minimum block size only possible in the form of a quart tree may be signaled from the encoding device 100 to the decoding device 200 through a bitstream. In addition, this information may be signaled for at least one unit of video, sequence, picture, parameter, tile group and slice (or segment).

Alternatively, the maximum block size and / or the minimum block size may be fixed sizes defined by the encoding device 100 and the decoding device 200. For example, when the size of a block is 64x64 or more, and 256x256 or less, only a quad tree type partition may be possible. In this case, split_flag may be a flag indicating whether to split the quad tree.

When the size of the block is larger than the maximum transform block size, only quad-tree type partitioning may be possible. At this time, the block to be divided may be at least one of a CU and a TU.

In this case, split_flag may be a flag indicating whether to split the quad tree.

If the size of a block falls within a specified range, only binary tree type or ternary tree type partitioning may be possible. Here, for example, the specified range may be defined by at least one of a maximum block size and a minimum block size, which are only possible in a binary tree or a ternary tree.

Information indicating the maximum block size and / or the minimum block size only possible in the form of a binary tree split or a ternary tree may be signaled from the encoding apparatus 100 to the decoding apparatus 200 through a bitstream. In addition, this information may be signaled for at least one unit of a sequence, picture and slice (or segment).

Alternatively, the maximum block size and / or the minimum block size may be fixed sizes defined by the encoding device 100 and the decoding device 200. For example, when the size of a block is 8x8 or more and 16x16 or less, only a binary tree-like partition may be possible. In this case, split_flag may be a flag indicating whether to split into a binary tree shape or a ternary tree shape.

The above description of the division of the code tree type may be equally applied to the division of the binary tree type and / or the ternary tree type.

Partitioning of blocks can be limited by previous partitioning. For example, when a block is divided into a specified binary tree form to generate a plurality of divided blocks, each divided block may be further divided only into a specified tree form. Here, the specified tree shape may be at least one of a binary tree shape, a ternary tree shape, and a quad tree shape.

If the horizontal size or vertical size of the divided block corresponds to a size that can no longer be divided, the above-described indicator may not be signaled.

7 is a diagram for explaining an embodiment of an intra prediction process.

Arrows from the center of the graph of FIG. 7 to the outside may indicate prediction directions of intra prediction modes. Also, a number displayed close to an arrow may represent an example of a mode value assigned to the prediction direction of the intra prediction mode or intra prediction mode.

Intra-coding and / or decoding may be performed using reference samples of neighboring units of the target block. The neighboring block may be a reconstructed neighboring block. The reference sample may mean a neighboring sample.

For example, intra encoding and / or decoding may be performed using a value or coding parameter of a reference sample included in a reconstructed neighboring block.

The encoding apparatus 100 and / or the decoding apparatus 200 may generate a prediction block by performing intra prediction on a target block based on information of a sample in the target image. When performing intra prediction, the encoding apparatus 100 and / or the decoding apparatus 200 may generate a prediction block for a target block by performing intra prediction based on information of a sample in the target image. When performing intra prediction, the encoding apparatus 100 and / or the decoding apparatus 200 may perform directional prediction and / or non-directional prediction based on at least one reconstructed reference sample.

The prediction block may mean a block generated as a result of performing intra prediction. The prediction block may correspond to at least one of CU, PU and TU.

The unit of the prediction block may be at least one of CU, PU and TU. The prediction block may have a square shape having a size of 2Nx2N or a size of NxN. The size of the NxN may include 4x4, 8x8, 16x16, 32x32 and 64x64.

Alternatively, the prediction block may be a square-shaped block having a size of 2x2, 4x4, 8x8, 16x16, 32x32 or 64x64, or may be a rectangular block having a size of 2x8, 4x8, 2x16, 4x16 and 8x16. have.

Intra prediction may be performed according to an intra prediction mode for a target block. The number of intra prediction modes that the target block may have may be a predetermined fixed value, or a value determined differently according to properties of the prediction block. For example, the properties of the prediction block may include the size of the prediction block and the type of the prediction block. In addition, the properties of the prediction block may indicate coding parameters for the prediction block.

For example, the number of intra prediction modes may be fixed to N regardless of the size of the prediction block. Or, for example, the number of intra prediction modes may be 3, 5, 9, 17, 34, 35, 36, 65 or 67.

The intra prediction mode may be a non-directional mode or a directional mode. For example, the intra prediction mode may include two non-directional modes and 33 directional modes as shown in FIG. 7.

The two non-directional modes may include a DC mode and a Planar mode.

The directional mode may be a prediction mode having a specific direction or a specific angle. The directional mode may be referred to as an argonular mode.

The intra prediction mode may be represented by at least one of a mode number, mode value mode angle, and mode direction. In other words, the terms "(mode) number of intra prediction mode", "(mode) value of intra prediction mode", "(mode) angle of intra prediction mode" and "(mode) direction of intra prediction mode) have the same meaning It can be used as, or can be used interchangeably.

The number of intra prediction modes may be M. M may be 1 or more. In other words, the intra prediction modes may be M numbers including the number of non-directional modes and the number of directional modes.

The number of intra prediction modes may be fixed to M regardless of the size and / or color component of the block. For example, the number of intra prediction modes can be fixed to either 35 or 67, regardless of the size of the block.

Alternatively, the number of intra prediction modes may be different depending on the size of the block and / or the type of color component.

For example, as the size of a block increases, the number of intra prediction modes may increase. Alternatively, as the size of the block increases, the number of intra prediction modes may decrease. If the block size is 4x4 or 8x8, the number of intra prediction modes may be 67. If the size of the block is 16x16, the number of intra prediction modes may be 35. When the block size is 32x32, the number of intra prediction modes may be 19. When the block size is 64x64, the number of intra prediction modes may be 7.

For example, the number of intra prediction modes may be different depending on whether the color component is a luma signal or a chroma signal. Alternatively, the number of intra prediction modes of the luma component block may be greater than the number of intra prediction modes of the chroma component block.

For example, in the vertical mode in which the mode value is 26, prediction may be performed in the vertical direction based on the pixel value of the reference sample. For example, in the horizontal mode in which the mode value is 10, prediction may be performed in the horizontal direction based on the pixel value of the reference sample.

Even in a directional mode other than the above-described mode, the encoding apparatus 100 and the decoding apparatus 200 may perform intra prediction on the target unit using a reference sample according to an angle corresponding to the directional mode.

The intra prediction mode located on the right side of the vertical mode may be referred to as a vertical-right mode. The intra prediction mode located at the bottom of the horizontal mode may be referred to as a horizontal-below mode. For example, in FIG. 7, intra prediction modes in which the mode value is one of 27, 28, 29, 30, 31, 32, 33, and 34 may be vertical right modes 613. The intra prediction modes whose mode value is one of 2, 3, 4, 5, 6, 7, 8, and 9 may be horizontal lower modes 616.

The non-directional mode may include a DC mode and a planar mode. For example, the mode value of DC mode may be 1. The mode value of the planner mode may be 0.

The directional mode may include an angular mode. A mode other than the DC mode and the planner mode among the plurality of intra prediction modes may be a directional mode.

When the intra prediction mode is the DC mode, a prediction block may be generated based on an average of pixel values of a plurality of reference samples. For example, a pixel value of a prediction block may be determined based on an average of pixel values of a plurality of reference samples.

The number of intra prediction modes described above and the mode value of each intra prediction mode may be exemplary only. The number of intra prediction modes described above and the mode value of each intra prediction mode may be defined differently according to embodiments, implementations, and / or needs.

In order to perform intra prediction on the target block, a step of checking whether samples included in the reconstructed neighboring block can be used as a reference sample of the target block may be performed. When there is a sample that is not available as a reference sample of the target block among samples of the neighboring block, a value generated by copying and / or interpolation using a sample value of at least one of the samples included in the reconstructed neighboring block It can be replaced with the sample value of a sample that cannot be used as a reference sample. If the value generated by copying and / or interpolation is replaced with the sample value of the sample, the sample can be used as a reference sample of the target block.

When intra prediction is used, a filter may be applied to at least one of a reference sample or a prediction sample based on at least one of an intra prediction mode and a size of a target block.

The type of filter applied to at least one of the reference sample or the prediction sample may be different according to at least one of an intra prediction mode of a target block, a size of the target block, and a shape of the target block. The type of filter may be classified according to one or more of the number of filter taps, the value of the filter coefficient, and the filter strength.

When the intra prediction mode is the planner mode, in generating a prediction block of the target block, according to a position in the prediction block of the prediction target sample, the upper reference sample of the target sample, the left reference sample of the target sample, and the upper right reference sample of the target block And a sample value of a prediction target sample may be generated using a weighted sum of the lower left reference sample of the target block.

When the intra prediction mode is the DC mode, in generating a prediction block of the target block, an average value of upper reference samples and left reference samples of the target block may be used. In addition, filtering using values of reference samples may be performed on specified rows or specified columns in the target block. The specified rows can be one or more top rows adjacent to the reference sample. The specified columns can be one or more left columns adjacent to the reference sample.

When the intra prediction mode is a directional mode, a prediction block may be generated using the upper reference sample, the left reference sample, the upper right reference sample, and / or the lower left reference sample of the target block.

Real-time interpolation may be performed to generate the above-described predictive sample.

The intra prediction mode of the target block may be predicted from the intra prediction mode of the neighboring block of the target block, and information used for prediction may be entropy encoded / decoded.

For example, if the intra prediction modes of the target block and the neighboring block are the same, it can be signaled that the intra prediction modes of the target block and the neighboring block are the same using a predefined flag.

For example, among the intra prediction modes of a plurality of neighboring blocks, an indicator indicating the same intra prediction mode as the intra prediction mode of the target block may be signaled.

If the intra prediction modes of the target block and the neighboring block are different, information of the intra prediction mode of the target block may be encoded and / or decoded using entropy encoding and / or decoding.

8 is a diagram for describing a reference sample used in the intra prediction process.

The reconstructed reference samples used for intra prediction of the target block include lower left reference samples, left reference samples, upper left corner reference samples, and upper reference samples And upper-right (above-right) reference samples.

For example, the left reference samples may refer to a reconstructed reference pixel adjacent to the left side of the target block. The top reference samples may refer to a reconstructed reference pixel adjacent to the top of the target block. The upper left corner reference sample may refer to a reconstructed reference pixel located in the upper left corner of the target block. Also, the lower left reference samples may refer to a reference sample located at the bottom of the left sample line among samples located on the same line as the left sample line composed of the left reference samples. The upper right reference samples may refer to reference samples located on the right side of the upper pixel line among samples located on the same line as the upper sample line composed of upper reference samples.

When the size of the target block is NxN, the lower left reference samples, the left reference samples, the upper reference samples, and the upper right reference samples may each be N pieces.

A prediction block may be generated through intra prediction on a target block. The generation of the prediction block may include determining values of pixels of the prediction block. The size of the target block and the prediction block may be the same.

The reference sample used for intra prediction of the target block may vary according to the intra prediction mode of the target block. The direction of the intra prediction mode may indicate a dependency relationship between reference samples and pixels of the prediction block. For example, the value of the specified reference sample can be used as the value of one or more specified pixels of the prediction block. In this case, the specified reference sample and the specified one or more pixels of the prediction block may be samples and pixels designated by a straight line in the direction of the intra prediction mode. In other words, the value of the specified reference sample can be copied to the value of the pixel located in the reverse direction of the intra prediction mode. Alternatively, the pixel value of the prediction block may be a value of a reference sample located in the direction of the intra prediction mode based on the position of the pixel.

For example, when the intra prediction mode of the target block is a vertical mode having a mode value of 26, upper reference samples may be used for intra prediction. When the intra prediction mode is a vertical mode, a pixel value of a prediction block may be a value of a reference sample vertically positioned based on the position of the pixel. Therefore, top reference samples that are adjacent to the top of the target block can be used for intra prediction. Also, the values of the pixels in one row of the prediction block may be the same as the values of the top reference samples.

For example, when the intra prediction mode of the target block is a horizontal mode having a mode value of 10, left reference samples may be used for intra prediction. When the intra prediction mode is a horizontal mode, a pixel value of a prediction block may be a value of a reference sample positioned horizontally left with respect to the pixel. Therefore, left reference samples adjacent to the left side of the target block can be used for intra prediction. Also, the values of the pixels in one column of the prediction block may be the same as the values of the left reference samples.

For example, when the mode value of the intra prediction mode of the target block is 18, at least some of the left reference samples, the upper left corner reference sample, and the at least some intra prediction of the upper reference samples may be used. When the mode value of the intra prediction mode is 18, the pixel value of the prediction block may be a value of a reference sample located diagonally on the upper left side based on the pixel.

In addition, when an intra prediction mode having a mode value of 27, 28, 29, 30, 31, 32, 33 or 34 is used, at least some of the upper right reference samples may be used for intra prediction.

In addition, when an intra prediction mode with mode values of 2, 3, 4, 5, 6, 7, 8 or 9 is used, at least some of the lower left reference samples may be used for intra prediction.

In addition, when an intra prediction mode in which the mode value is one of 11 to 25 is used, the upper left corner reference sample may be used for intra prediction.

One reference sample used to determine the pixel value of one pixel of the prediction block may be one, or may be two or more.

As described above, the pixel value of the pixel of the prediction block may be determined according to the position of the pixel and the position of the reference sample indicated by the direction of the intra prediction mode. When the position of the pixel and the position of the reference sample indicated by the direction of the intra prediction mode are integer positions, the value of one reference sample indicated by the integer position may be used to determine the pixel value of the pixel of the prediction block.

When the position of the pixel and the position of the reference sample indicated by the direction of the intra prediction mode are not integer positions, an interpolated reference sample may be generated based on the two reference samples closest to the position of the reference sample. have. The value of the interpolated reference sample can be used to determine the pixel value of the pixel of the prediction block. In other words, when the position of the pixel of the prediction block and the position of the reference sample indicated by the direction of the intra prediction mode indicate between the two reference samples, an interpolated value is generated based on the values of the two samples above. Can be.

The prediction block generated by prediction may not be the same as the original target block. That is, a prediction error, which is a difference between a target block and a prediction block, may exist, and a prediction error may also exist between a pixel of the target block and a pixel of the prediction block.

Hereinafter, the terms "difference", "error" and "residual" may be used in the same sense and may be used interchangeably.

For example, in the case of directional intra prediction, a larger prediction distance may occur as the distance between a pixel of a prediction block and a reference sample increases. Discontinuity may occur between the prediction block and neighboring blocks generated by the prediction error.

Filtering on the prediction block can be used to reduce the prediction error. Filtering may be to adaptively apply a filter to a region considered to have a large prediction error among prediction blocks. For example, an area considered to have a large prediction error may be a boundary of a prediction block. Also, according to the intra prediction mode, a region considered as having a large prediction error among prediction blocks may be different, and characteristics of a filter may be different.

As illustrated in FIG. 8, for intra prediction of a target block, at least one of reference lines 0 to 3 may be used. Each reference line can represent a reference sample line. The smaller the reference line number, the more likely it may be a line of reference samples closer to the target block.

Samples of segments A and F can be obtained through padding using the closest samples of segments B and E, respectively, instead of being obtained from the reconstructed neighboring blocks.

Index information indicating a reference sample line to be used for intra prediction of a target block may be signaled. The index information may refer to a reference sample line used for intra prediction of a target block among a plurality of reference sample lines. For example, the index information may have one of 0 to 3.

When the upper boundary of the target block is the boundary of the CTU, only the reference sample line 0 is available. Therefore, in this case, index information may not be signaled. When a reference sample line other than the reference sample line 0 is used, filtering on a prediction block described below may not be performed.

In the case of inter-color intra prediction, a prediction block for a target block of the second color component may be generated based on a corresponding reconstructed block of the first color component.

For example, the first color component may be a luma component, and the second color component may be a chroma component.

For intra prediction between color components, parameters of the linear model between the first color component and the second color component can be derived based on the template.

The template may include a top reference sample and / or a left reference sample of the target block, and a top reference sample and / or a left reference sample of the reconstructed block of the first color component corresponding to these reference samples. have.

For example, the parameters of the linear model are 1) the value of the sample of the first color component having the maximum value among the samples in the template, 2) the value of the sample of the second color component corresponding to the sample of the first color component, 3) a sample value of a first color component having a minimum value among samples in a template and 4) a value of a sample of a second color component corresponding to a sample of the first color component.

When the parameters of the linear model are derived, a predictive block for the target block can be generated by applying the corresponding reconstructed block to the linear model.

Depending on the image format, sub-sampling may be performed on neighboring samples of the reconstructed blocks of the first color component and corresponding reconstructed blocks. For example, when one sample of the second color component corresponds to four samples of the first color component, one corresponding sample can be calculated by subsampling the four samples of the first color component. have. When sub-sampling is performed, derivation of parameters of the linear model and intra prediction between color components may be performed based on the sub-sampled corresponding sample.

Whether intra prediction between color components is performed and / or a range of templates may be signaled as an intra prediction mode.

The target block may be divided into two or four sub-blocks in the horizontal direction and / or the vertical direction.

The divided sub-blocks may be reconstructed sequentially. That is, as intra prediction is performed on the sub-block, a sub-prediction block for the sub-block may be generated. In addition, as inverse quantization and / or inverse transformation is performed on the subblock, a sub residual block for the subblock may be generated. A reconstructed subblock may be generated by adding the sub prediction block to the sub residual block. The reconstructed sub-block may be used as a reference sample for intra prediction of a sub-block of a higher rank.

The sub-block may be a block including a specified number (eg, 16) or more samples. Thus, for example, when the target block is an 8x4 block or a 4x8 block, the target block may be divided into two sub-blocks. Also, when the target block is a 4x4 block, the target block cannot be divided into sub-blocks. When the target block has a different size, the target block may be divided into four sub-blocks.

Information regarding whether intra prediction based on the sub-block is performed and / or the split direction (horizontal or vertical) may be signaled.

Such sub-block-based intra prediction may be limited to be performed only when the reference sample line 0 is used. When sub-block-based intra prediction is performed, filtering on a prediction block described below may not be performed.

The final prediction block may be generated by performing filtering on the prediction block generated by intra prediction.

Filtering may be performed by applying a specific weight to a filtering target sample, a left reference sample, a top reference sample, and / or a top left reference sample, which are the targets of filtering.

The weight used for filtering and / or the reference sample (or the range of the reference sample or the position of the reference sample, etc.) may be determined based on at least one of block size, intra prediction mode, and position within the prediction block of the sample to be filtered have.

For example, filtering can be performed only for a specified intra prediction mode (eg, DC mode, planner mode, vertical mode, horizontal mode, diagonal mode, and / or adjacent diagonal mode).

The adjacent diagonal mode may be a mode having a number of k added to the number of the diagonal mode, or a mode having a number of k subtracted from the number of the diagonal mode. That is, the number of adjacent diagonal modes may be the sum of the numbers and k of the diagonal modes, and may be the difference between the numbers and k of the diagonal modes. For example, k may be a positive integer of 8 or less.

The intra prediction mode of the target block may be derived using an intra prediction mode of neighboring blocks existing around the target block, and the derived intra prediction mode may be entropy encoded and / or entropy decoded.

For example, if the intra prediction mode of the target block and the intra prediction mode of the neighboring block are the same, information that the intra prediction mode of the target block and the intra prediction mode of the neighboring block are the same may be signaled using the specified flag information. .

Further, for example, indicator information for a neighboring block having the same intra prediction mode as the intra prediction mode of the target block among intra prediction modes of a plurality of neighboring blocks may be signaled.

For example, if the intra prediction mode of the target block and the intra prediction mode of the neighboring block are different, information about the intra prediction mode of the target block is performed by performing entropy encoding and / or entropy decoding based on the intra prediction mode of the neighboring block. Entropy encoding and / or entropy decoding may be performed.

9 is a diagram for explaining an embodiment of an inter prediction process.

The square illustrated in FIG. 9 may represent an image (or picture). Also, in FIG. 9, an arrow may indicate a prediction direction. An arrow from the first picture to the second picture may indicate that the second picture refers to the first picture. That is, the image may be encoded and / or decoded according to the prediction direction.

Each image may be classified into an I picture (Intra Picture), a P picture (Uni-prediction Picture), and a B picture (Bi-prediction Picture) according to an encoding type. Each picture may be encoded and / or decoded according to the encoding type of each picture.

When the target image to be encoded is an I picture, the target image may be encoded using data in the image itself without inter prediction referring to another image. For example, the I picture can be encoded only with intra prediction.

When the target image is a P picture, the target image may be coded through inter prediction using only reference pictures existing in one direction. Here, the unidirectional may be forward or reverse.

When the target image is a B picture, the target image may be encoded through inter prediction using reference pictures existing in both directions or inter prediction using a reference picture existing in one of forward and backward directions. Here, both directions may be forward and backward.

P pictures and B pictures that are encoded and / or decoded using a reference picture may be regarded as an image in which inter prediction is used.

Hereinafter, inter prediction in the inter mode according to an embodiment is described in detail.

Inter prediction or motion compensation may be performed using reference images and motion information.

In the inter mode, the encoding apparatus 100 may perform inter prediction and / or motion compensation on the target block. The decoding apparatus 200 may perform inter prediction and / or motion compensation corresponding to inter prediction and / or motion compensation in the encoding apparatus 100 on the target block.

Motion information on the target block may be derived during inter prediction by each of the encoding device 100 and the decoding device 200. Motion information may be derived using motion information of a reconstructed neighboring block, motion information of a call block, and / or motion information of a block adjacent to a call block.

For example, the encoding apparatus 100 or the decoding apparatus 200 performs prediction and / or motion compensation by using motion information of a spatial candidate and / or a temporal candidate as motion information of a target block. It can be done. The target block may mean PU and / or PU partition.

The spatial candidate may be a reconstructed block spatially adjacent to the target block.

The temporal candidate may be a reconstructed block corresponding to a target block in a collocated picture (collocated picture).

In inter prediction, the encoding apparatus 100 and the decoding apparatus 200 may improve encoding efficiency and decoding efficiency by using motion information of spatial candidates and / or temporal candidates. The motion information of the spatial candidate may be referred to as spatial motion information. The motion information of the temporal candidate may be referred to as temporal motion information.

Hereinafter, the motion information of the spatial candidate may be motion information of the PU including the spatial candidate. The motion information of the temporal candidate may be motion information of the PU including the temporal candidate. The motion information of the candidate block may be motion information of a PU including the candidate block.

Inter prediction may be performed using a reference picture.

The reference picture may be at least one of a previous picture of the target picture or a subsequent picture of the target picture. The reference picture may mean an image used for prediction of a target block.

In inter prediction, a region within a reference picture may be specified by using a reference picture index (or refIdx) indicating a reference picture and a motion vector to be described later. Here, the specified region in the reference picture may represent a reference block.

For inter prediction, a reference picture may be selected, and a reference block corresponding to a target block may be selected in the reference picture. Also, inter prediction may generate a prediction block for a target block using the selected reference block.

Motion information may be derived during inter prediction by each of the encoding device 100 and the decoding device 200.

The spatial candidate may be 1) existing in the target picture, 2) already reconstructed through encoding and / or decoding, 3) adjacent to the target block, or a block located at a corner of the target block. Here, the block located at the corner of the target block may be a block vertically adjacent to a neighboring block horizontally adjacent to the target block or a block horizontally adjacent to a neighboring block vertically adjacent to the target block. The "block located at the corner of the target block" may have the same meaning as "block adjacent to the corner of the target block". The "block located at the corner of the target block" may be included in the "block adjacent to the target block".

For example, the spatial candidate is a reconstructed block located on the left side of the target block, a reconstructed block located on the top of the target block, a reconstructed block located on the lower left corner of the target block, and located on the upper right corner of the target block. It may be a rebuilt block or a rebuilt block located in the upper left corner of the target block.

Each of the encoding apparatus 100 and the decoding apparatus 200 may identify a block existing in a position spatially corresponding to a target block in a call picture. The position of the target block in the target picture and the position of the identified block in the call picture may correspond to each other.

Each of the encoding apparatus 100 and the decoding apparatus 200 may determine, as a temporal candidate, a call block existing at a predetermined relative position with respect to the identified block. The predefined relative position may be an internal position and / or an external position of the identified block.

For example, the call block may include a first call block and a second call block. When the coordinates of the identified block are (xP, yP), and the size of the identified block is (nPSW, nPSH), the first call block may be a block located at coordinates (xP + nPSW, yP + nPSH). The second call block may be a block located at coordinates (xP + (nPSW >> 1), yP + (nPSH >> 1)). The second call block may be selectively used when the first call block is unavailable.

The motion vector of the target block may be determined based on the motion vector of the call block. Each of the encoding device 100 and the decoding device 200 may scale a motion vector of a call block. The scaled motion vector of the call block can be used as the motion vector of the target block. Also, the motion vector of the motion information of the temporal candidate stored in the list may be a scaled motion vector.

The ratio of the motion vector of the target block and the motion vector of the call block may be equal to the ratio of the first temporal distance and the second temporal distance. The first temporal distance may be a distance between a reference picture of a target block and a target picture. The second temporal distance may be a distance between a reference picture and a call picture of a call block.

The method of deriving motion information may vary according to the inter prediction mode of the target block. For example, as an inter prediction mode applied for inter prediction, an advanced motion vector predictor (AMVP) mode, a merge mode and a skip mode, and a merge mode having a motion vector difference, There may be a sub-block merge mode, a triangulation mode, an inter-intra prediction mode, an affine inter mode, and a current picture reference mode. The merge mode may be referred to as a motion merge mode. In the following, each of the modes is described in detail.

1) AMVP mode

When the AMVP mode is used, the encoding apparatus 100 may search for similar blocks in the neighborhood of the target block. The encoding apparatus 100 may obtain a prediction block by performing prediction on the target block using motion information of the searched similar block. The encoding apparatus 100 may encode a residual block that is a difference between a target block and a prediction block.

1-1) Preparation of a prediction motion vector candidate list

When the AMVP mode is used as the prediction mode, each of the encoding apparatus 100 and the decoding apparatus 200 can generate a prediction motion vector candidate list using a motion vector of a spatial candidate, a motion vector of a temporal candidate, and a zero vector. have. The predicted motion vector candidate list may include one or more predicted motion vector candidates. At least one of a spatial candidate motion vector, a temporal candidate motion vector, and a zero vector may be determined and used as a predictive motion vector candidate.

Hereinafter, the terms “prediction motion vector (candidate)” and “motion vector (candidate)” may be used in the same sense and may be used interchangeably.

Hereinafter, the terms “prediction motion vector candidate” and “AMVP candidate” may be used in the same sense and may be used interchangeably.

Hereinafter, the terms “prediction motion vector candidate list” and “AMVP candidate list” may be used in the same sense, and may be used interchangeably.

The spatial candidate may include a reconstructed spatial neighboring block. In other words, the motion vector of the reconstructed neighboring block may be referred to as a spatial prediction motion vector candidate.

The temporal candidate may include a call block and a block adjacent to the call block. In other words, a motion vector of a call block or a motion vector of a block adjacent to a call block may be referred to as a temporal prediction motion vector candidate.

The zero vector may be a (0, 0) motion vector.

The predicted motion vector candidate may be a motion vector predictor for prediction of a motion vector. Also, in the encoding apparatus 100, a predicted motion vector candidate may be a motion vector initial search position.

1-2) Searching for motion vectors using predicted motion vector candidate list

The encoding apparatus 100 may use the predicted motion vector candidate list to determine a motion vector to be used for encoding a target block within a search range. Also, the encoding apparatus 100 may determine a prediction motion vector candidate to be used as a prediction motion vector of a target block among prediction motion vector candidates in the prediction motion vector candidate list.

The motion vector to be used for encoding the target block may be a motion vector that can be encoded with minimal cost.

In addition, the encoding apparatus 100 may determine whether to use the AMVP mode in encoding the target block.

1-3) Transmission of inter prediction information

The encoding apparatus 100 may generate a bitstream including inter prediction information required for inter prediction. The decoding apparatus 200 may perform inter prediction on a target block using inter prediction information of a bitstream.

The inter prediction information includes 1) mode information indicating whether to use the AMVP mode, 2) predicted motion vector index, 3) motion vector difference (MVD), 4) reference direction, and 5) reference picture index. can do.

Hereinafter, the terms "prediction motion vector index" and "AMVP index" may be used with the same meaning, and may be used interchangeably.

Also, the inter prediction information may include a residual signal.

When the mode information indicates that the AMVP mode is used, the decoding apparatus 200 may obtain a predicted motion vector index, motion vector difference, reference direction, and reference picture index from the bitstream through entropy decoding.

The predicted motion vector index may indicate a predicted motion vector candidate used for prediction of a target block among predicted motion vector candidates included in the predicted motion vector candidate list.

1-4) Inter prediction in AMVP mode using inter prediction information

The decoding apparatus 200 may derive a predicted motion vector candidate using a predicted motion vector candidate list, and determine motion information of a target block based on the derived predicted motion vector candidate.

The decoding apparatus 200 may determine a motion vector candidate for the target block from among the predicted motion vector candidates included in the predicted motion vector candidate list using the predicted motion vector index. The decoding apparatus 200 may select a prediction motion vector candidate indicated by the prediction motion vector index from among prediction motion vector candidates included in the prediction motion vector candidate list, as a prediction motion vector of the target block.

The encoding apparatus 100 may generate an entropy-encoded prediction motion vector index by applying entropy encoding to the prediction motion vector index, and may generate a bitstream including the entropy-encoded prediction motion vector index. The entropy-encoded predicted motion vector index may be signaled from the encoding apparatus 100 to the decoding apparatus 200 through a bitstream. The decoding apparatus 200 may extract an entropy-encoded predicted motion vector index from the bitstream, and obtain a predicted motion vector index by applying entropy decoding to the entropy-encoded predicted motion vector index.

The motion vector to be actually used for inter prediction of the target block may not match the predicted motion vector. A motion vector to be actually used for inter prediction of a target block and MVD may be used to indicate a difference between the predicted motion vectors. The encoding apparatus 100 may derive a predicted motion vector similar to a motion vector to be actually used for inter prediction of a target block in order to use MVD having a size as small as possible.

MVD may be a difference between a motion vector and a predicted motion vector of a target block. The encoding apparatus 100 may calculate an MVD, and generate entropy-encoded MVD by applying entropy encoding to the MVD. The encoding apparatus 100 may generate a bitstream including an entropy-encoded MDV.

The MVD may be transmitted from the encoding device 100 to the decoding device 200 through a bitstream. The decoding apparatus 200 may extract the entropy-encoded MVD from the bitstream, and obtain the MVD by applying entropy decoding to the entropy-encoded MVD.

The decoding apparatus 200 may derive the motion vector of the target block by adding the MVD and the predicted motion vector. In other words, the motion vector of the target block derived from the decoding apparatus 200 may be a sum of MVD and motion vector candidates.

In addition, the encoding apparatus 100 may generate entropy-encoded MVD resolution information by applying entropy encoding to the calculated MVD resolution information, and may generate a bitstream including entropy-encoded MVD resolution information. The decoding apparatus 200 may extract entropy-encoded MVD resolution information from a bitstream, and obtain MVD resolution information by applying entropy decoding to entropy-encoded MVD resolution information. The decoding apparatus 200 may adjust the resolution of the MVD using the MVD resolution information.

Meanwhile, the encoding apparatus 100 may calculate MVD based on an affine model. The decoding apparatus 200 may derive the affine control motion vector of the target block through the sum of the MVD and affine control motion vector candidates, and may derive a motion vector for the subblock using the affine control motion vector. have.

The reference direction may refer to a list of reference pictures used for prediction of a target block. For example, the reference direction may point to one of the reference picture list L0 and the reference picture list L1.

The reference direction only points to a list of reference pictures used for prediction of a target block, and may not indicate that the directions of reference pictures are limited to a forward direction or a backward direction. That is, each of the reference picture list L0 and the reference picture list L1 may include pictures in the forward and / or reverse directions.

When the reference direction is uni-direction, it may mean that one reference picture list is used. When the reference direction is bi-direction, it may mean that two reference picture lists are used. In other words, the reference direction may indicate that only the reference picture list L0 is used, only the reference picture list L1 is used, and one of the two reference picture lists.

The reference picture index may indicate a reference picture used for prediction of a target block among reference pictures in a reference picture list. The encoding apparatus 100 may generate an entropy-encoded reference picture index by applying entropy encoding to the reference picture index, and may generate a bitstream including the entropy-encoded reference picture index. The entropy-encoded reference picture index may be signaled from the encoding device 100 to the decoding device 200 through a bitstream. The decoding apparatus 200 may extract an entropy-encoded reference picture index from the bitstream, and obtain a reference picture index by applying entropy decoding to the entropy-encoded reference picture index.

When two reference picture lists are used for prediction of a target block. One reference picture index and one motion vector may be used for each reference picture list. In addition, when two reference picture lists are used for prediction of a target block, two prediction blocks may be specified for the target block. For example, a (final) prediction block of the target block may be generated through an average or weighted-sum of two prediction blocks for the target block.

The motion vector of the target block may be derived by the predicted motion vector index, MVD, reference direction, and reference picture index.

The decoding apparatus 200 may generate a prediction block for the target block based on the derived motion vector and reference picture index. For example, the prediction block may be a reference block indicated by a derived motion vector in the reference picture indicated by the reference picture index.

By encoding the predicted motion vector index and MVD without encoding the motion vector itself of the target block, the amount of bits transmitted from the encoding apparatus 100 to the decoding apparatus 200 may be reduced, and encoding efficiency may be improved.

Motion information of a neighbor block reconstructed with respect to the target block may be used. In a specific inter prediction mode, the encoding apparatus 100 may not separately encode motion information itself for a target block. The motion information of the target block is not encoded, and other information capable of deriving the motion information of the target block through motion information of the reconstructed neighboring block may be encoded instead. As other information is encoded instead, the amount of bits transmitted to the decoding apparatus 200 may be reduced, and encoding efficiency may be improved.

For example, as an inter prediction mode in which motion information of the target block is not directly encoded, there may be a skip mode and / or a merge mode. In this case, the encoding apparatus 100 and the decoding apparatus 200 may use an identifier and / or an index indicating which of the reconstructed neighboring units is used as motion information of the target unit.

2) Merge mode

As a method of deriving motion information of a target block, there is a merge. Merge may mean merging of motions for a plurality of blocks. Merge may mean applying motion information of one block to another block together. In other words, the merge mode may mean a mode in which motion information of a target block is derived from motion information of a neighboring block.

When the merge mode is used, the encoding apparatus 100 may predict motion information of a target block using motion information of a spatial candidate and / or motion information of a temporal candidate. The spatial candidate may include a reconstructed spatial neighboring block spatially adjacent to the target block. The spatial neighboring block may include a left neighboring block and an upper neighboring block. The temporal candidate may include a call block. The terms "spatial candidate" and "spatial merge candidate" may be used interchangeably and may be used interchangeably. The terms "temporal candidate" and "temporal merge candidate" may be used interchangeably and may be used interchangeably.

The encoding apparatus 100 may obtain a prediction block through prediction. The encoding apparatus 100 may encode a residual block that is a difference between a target block and a prediction block.

2-1) Preparation of a merge candidate list

When the merge mode is used, each of the encoding apparatus 100 and the decoding apparatus 200 may generate a merge candidate list using motion information of spatial candidates and / or motion information of temporal candidates. The motion information may include 1) a motion vector, 2) a reference picture index, and 3) a reference direction. The reference direction can be unidirectional or bidirectional. The reference direction may mean an inter prediction indicator.

The merge candidate list may include merge candidates. The merge candidate may be motion information. In other words, the merge candidate list may be a list in which motion information is stored.

The merge candidates may be motion information such as temporal candidates and / or spatial candidates. In other words, the merge candidate list may include motion information such as temporal candidates and / or spatial candidates.

Further, the merge candidate list may include a new merge candidate generated by a combination of merge candidates already existing in the merge candidate list. In other words, the merge candidate list may include new motion information generated by a combination of motion information already existing in the merge candidate list.

Also, the merge candidate list may include a history-based merge candidate. The history-based merge candidate may be motion information of a block that is encoded and / or decoded before the target block.

The merge candidates may be specified modes for deriving inter prediction information. The merge candidate may be information indicating a specific mode for deriving inter prediction information. Inter prediction information of the target block may be derived according to the specified mode indicated by the merge candidate. At this time, the specified mode may include a process of deriving a series of inter prediction information. The specified mode may be an inter prediction information derivation mode or a motion information derivation mode.

Inter prediction information of a target block may be derived according to a mode indicated by a merge candidate selected by a merge index among merge candidates in a merge candidate list.

For example, the motion information derivation modes in the merge candidate list may be at least one of 1) a sub-block unit motion information derivation mode and 2) affine motion information derivation mode.

Also, the merge candidate list may include motion information of a zero vector. The zero vector may be referred to as a zero merge candidate.

In other words, motion information in the merge candidate list includes: 1) spatial candidate motion information, 2) temporal candidate motion information, 3) motion information generated by a combination of motion information already present in the merge candidate list, 4) zero vector. It may be at least one of.

The motion information may include 1) a motion vector, 2) a reference picture index, and 3) a reference direction. The reference direction may be referred to as an inter prediction indicator. The reference direction can be unidirectional or bidirectional. The unidirectional reference direction may indicate L0 prediction or L1 prediction.

The merge candidate list may be generated before prediction by the merge mode is performed.

The number of merge candidates in the merge candidate list may be predefined. The encoding apparatus 100 and the decoding apparatus 200 may add a merge candidate to the merge candidate list according to a predefined method and a predefined rank so that the merge candidate list has a predefined number of merge candidates. The merge candidate list of the encoding apparatus 100 and the merge candidate list of the decoding apparatus 200 may be the same through a predefined method and a predefined ranking.

Merge may be applied in CU units or PU units. When merging is performed in a CU unit or a PU unit, the encoding device 100 may transmit a bitstream including predefined information to the decoding device 200. For example, the predefined information includes: 1) information indicating whether to perform merging for each block partition, 2) which block to merge with which of the spatial candidate and / or temporal candidate blocks for the target block Information.

2-2) Searching for motion vectors using merge candidate lists

The encoding apparatus 100 may determine a merge candidate to be used for encoding the target block. For example, the encoding apparatus 100 may perform predictions on the target block using merge candidates of the merge candidate list, and generate residual blocks for merge candidates. The encoding apparatus 100 may use merge candidates that require a minimum cost in encoding prediction and residual blocks for encoding the target block.

Also, the encoding apparatus 100 may determine whether to use the merge mode in encoding the target block.

2-3) Transmission of inter prediction information

The encoding apparatus 100 may generate a bitstream including inter prediction information required for inter prediction. The encoding apparatus 100 may generate entropy-encoded inter-prediction information by performing entropy encoding on the inter-prediction information, and may transmit a bitstream including entropy-encoded inter-prediction information to the decoding apparatus 200. Through the bitstream, entropy-encoded inter prediction information may be signaled from the encoding apparatus 100 to the decoding apparatus 200. The decoding apparatus 200 may extract entropy-encoded inter-prediction information from the bitstream, and may obtain inter-prediction information by performing entropy decoding on the entropy-encoded inter-prediction information.

The decoding apparatus 200 may perform inter prediction on a target block using inter prediction information of a bitstream.

The inter prediction information may include 1) mode information indicating whether to use the merge mode, 2) merge index, and 3) correction information.

Also, the inter prediction information may include a residual signal.

The decoding apparatus 200 may acquire the merge index from the bitstream only when the mode information indicates that the merge mode is used.

The mode information may be a merge flag. The unit of mode information may be a block. Information about the block may include mode information, and the mode information may indicate whether a merge mode is applied to the block.

The merge index may indicate a merge candidate used for prediction of a target block among merge candidates included in the merge candidate list. Alternatively, the merge index may indicate which of the neighboring blocks spatially or temporally adjacent to the target block is merged.

The encoding apparatus 100 may select a merge candidate having the highest encoding performance among merge candidates included in the merge candidate list, and may set a value of the merge index to indicate the selected merge candidate.

The correction information may be information used for correction of a motion vector. The encoding apparatus 100 may generate correction information. The decoding apparatus 200 may correct the motion vector of the merge candidate selected by the merge index based on the correction information.

The correction information may include at least one of information indicating whether correction is performed, correction direction information, and correction size information. A prediction mode for correcting a motion vector based on signaled correction information may be referred to as a merge mode having a motion vector difference.

2-4) Inter prediction in merge mode using inter prediction information

The decoding apparatus 200 may perform prediction for the target block by using the merge candidate indicated by the merge index among merge candidates included in the merge candidate list.

The motion vector of the target block may be specified by a motion vector, a reference picture index, and a reference direction of the merge candidate indicated by the merge index.

3) Skip mode

The skip mode may be a mode in which motion information of a spatial candidate or motion information of a temporal candidate is applied to the target block as it is. Also, the skip mode may be a mode that does not use a residual signal. That is, when the skip mode is used, the reconstructed block may be the same as the prediction block.

The difference between the merge mode and the skip mode may be whether to transmit or use the residual signal. In other words, the skip mode can be similar to the merge mode, except that the residual signal is not transmitted or used.

When the skip mode is used, the encoding apparatus 100 transmits information indicating which block motion information among spatial candidates or temporal candidate blocks is used as motion information of a target block to the decoding apparatus 200 through a bitstream. Can transmit. The encoding apparatus 100 may generate entropy-encoded information by performing entropy encoding on such information, and may signal entropy-encoded information to the decoding apparatus 200 through a bitstream. The decoding apparatus 200 may extract entropy-encoded information from the bitstream, and obtain information by performing entropy decoding on the entropy-encoded information.

Also, when the skip mode is used, the encoding apparatus 100 may not transmit other syntax element information such as MVD to the decoding apparatus 200. For example, when the skip mode is used, the encoding apparatus 100 may not signal syntax elements related to at least one of MVD, coded block flags, and transform coefficient levels to the decoding apparatus 200.

3-1) Preparation of merge candidate list

The skip mode may also use a merge candidate list. In other words, the merge candidate list can be used in both merge mode and skip mode. In this aspect, the merge candidate list may be referred to as a "skip candidate list" or a "merge / skip candidate list".

Alternatively, the skip mode may use a separate candidate list different from the merge mode. In this case, in the description below, the merge candidate list and the merge candidate may be replaced with a skip candidate list and a skip candidate, respectively.

The merge candidate list may be generated before prediction by the skip mode is performed.

3-2) Searching motion vectors using merge candidate list

The encoding apparatus 100 may determine a merge candidate to be used for encoding the target block. For example, the encoding apparatus 100 may perform predictions for the target block using merge candidates of the merge candidate list. The encoding apparatus 100 may use a merge candidate that requires a minimum cost in prediction for encoding the target block.

Also, the encoding apparatus 100 may determine whether to use the skip mode in encoding the target block.

3-3) Transmission of inter prediction information

The encoding apparatus 100 may generate a bitstream including inter prediction information required for inter prediction. The decoding apparatus 200 may perform inter prediction on a target block using inter prediction information of a bitstream.

The inter prediction information may include 1) mode information indicating whether to use the skip mode and 2) skip index.

The skip index may be the same as the merge index described above.

When the skip mode is used, the target block can be coded without a residual signal. The inter prediction information may not include a residual signal. Or, the bitstream may not include a residual signal.

The decoding apparatus 200 may acquire the skip index from the bitstream only when the mode information indicates that the skip mode is used. As described above, the merge index and skip index may be the same. The decoding apparatus 200 may acquire the skip index from the bitstream only when the mode information indicates that the merge mode or the skip mode is used.

The skip index may indicate a merge candidate used for prediction of a target block among merge candidates included in the merge candidate list.

3-4) Inter prediction of skip mode using inter prediction information

The decoding apparatus 200 may perform prediction for the target block by using the merge candidate indicated by the skip index among merge candidates included in the merge candidate list.

The motion vector of the target block may be specified by the motion vector, reference picture index, and reference direction of the merge candidate indicated by the skip index.

4) Current picture reference mode

The current picture reference mode may refer to a prediction mode using a pre-reconstructed region in the target picture to which the target block belongs.

Motion vectors to specify pre-reconstructed regions can be used. Whether the target block is encoded in the current picture reference mode may be determined using a reference picture index of the target block.

A flag or an index indicating whether the target block is a block encoded in the current picture reference mode may be signaled from the encoding device 100 to the decoding device 200. Alternatively, whether the target block is a block encoded in the current picture reference mode may be inferred through a reference picture index of the target block.

When the target block is encoded in the current picture reference mode, the target picture may exist at a fixed position or an arbitrary position in the reference picture list for the target block.

For example, the fixed position may be a position where the value of the reference picture index is 0 or the last position.

When the target picture is present at an arbitrary position in the reference picture list, a separate reference picture index indicating this arbitrary position may be signaled from the encoding device 100 to the decoding device 200.

5) subblock merge mode

The sub-block merge mode may refer to a mode for deriving motion information on a sub-block of the CU.

When a sub-block merge mode is applied, motion information of a call sub-block of a target sub-block (in other words, a sub-block based temporal merge candidate) and / or an affine control point motion vector in a reference image A subblock merge candidate list may be generated using a merge control point motion vector merge candidate.

6) triangular partition mode

In the triangular segmentation mode, divided object blocks may be generated by dividing the object block in a diagonal direction. For each divided target block, motion information of each divided target block may be derived, and a predicted sample for each divided target block may be derived using the derived motion information. A predicted sample of the target block may be derived through a weighted sum of the predicted samples of the partitioned target blocks.

7) Inter intra combining prediction mode

The inter-intra prediction mode may be a mode in which a prediction sample of a target block is derived using a weighted sum of a prediction sample generated by inter prediction and a prediction sample generated by intra prediction.

In the above-described modes, the decoding apparatus 200 may perform its own correction for the derived motion information. For example, the decoding apparatus 200 may search for a specific region based on a reference block indicated by the derived motion information and search for motion information having a minimum sum of absolute differences (SAD). The retrieved motion information can be derived as corrected motion information.

In the above-described modes, the decoding apparatus 200 may compensate for a prediction sample derived through inter prediction using an optical flow.

In the above-described AMVP mode, merge mode, and skip mode, motion information to be used for prediction of a target block among motion information in a list may be specified through an index for a list.

In order to improve encoding efficiency, the encoding apparatus 100 may signal only the index of an element that causes a minimum cost in inter prediction of a target block among elements of a list. The encoding apparatus 100 may encode an index and signal the encoded index.

Accordingly, the above-described lists (that is, the predicted motion vector candidate list and the merge candidate list) may be derived in the same manner based on the same data in the encoding apparatus 100 and the decoding apparatus 200. Here, the same data may include a reconstructed picture and a reconstructed block. Also, in order to specify elements by index, the order of elements in the list may need to be constant.

10 shows spatial candidates according to an example.

In Figure 10, the location of spatial candidates is shown.

The large block in the middle may represent the target block. The five small blocks can represent spatial candidates.

The coordinates of the target block may be (xP, yP), and the size of the target block may be (nPSW, nPSH).

The spatial candidate A ₀ may be a block adjacent to the lower left corner of the target block. A ₀ may be a block occupying a pixel of coordinates (xP-1, yP + nPSH + 1).

The spatial candidate A ₁ may be a block adjacent to the left side of the target block. A ₁ may be the lowest block among blocks adjacent to the left side of the target block. Alternatively, A ₁ may be a block adjacent to the top of A ₀ . A ₁ may be a block occupying a pixel of coordinates (xP-1, yP + nPSH).

The spatial candidate B ₀ may be a block adjacent to the upper right corner of the target block. B ₀ may be a block occupying a pixel of coordinates (xP + nPSW + 1, yP-1).

The spatial candidate B ₁ may be a block adjacent to the top of the target block. B ₁ may be the rightmost block among blocks adjacent to the top of the target block. Alternatively, B ₁ may be a block adjacent to the left side of B ₀ . B ₁ may be a block occupying a pixel of coordinates (xP + nPSW, yP-1).

The spatial candidate B ₂ may be a block adjacent to the upper left corner of the target block. B ₂ may be a block occupying a pixel of coordinates (xP-1, yP-1).

Determination of availability of spatial and temporal candidates

In order to include the motion information of the spatial candidate or the motion information of the temporal candidate in the list, it is necessary to determine whether the spatial candidate motion information or the temporal candidate motion information is available.

Hereinafter, the candidate block may include spatial candidates and temporal candidates.

For example, the above determination can be made by sequentially applying steps 1) to 4) below.

Step 1) If the PU including the candidate block is outside the boundary of the picture, the availability of the candidate block may be set to false. “Availability set to false” may have the same meaning as “set to unavailable”.

Step 2) If the PU including the candidate block is outside the boundary of the slice, the availability of the candidate block may be set to false. If the target block and the candidate block are located in different slices, the availability of the candidate block can be set to false.

Step 3) If the PU including the candidate block is outside the tile boundary, the availability of the candidate block may be set to false. If the target block and the candidate block are located in different tiles, the availability of the candidate block can be set to false.

Step 4) If the prediction mode of the PU including the candidate block is an intra prediction mode, the availability of the candidate block may be set to false. If the PU including the candidate block does not use inter prediction, the availability of the candidate block may be set to false.

11 shows an additional order of motion information of spatial candidates according to an example to a merge list.

As illustrated in FIG. 11, in adding motion information of spatial candidates to the merge list, the order of A ₁ , B ₁ , B ₀ , A ₀ and B ₂ can be used. That is, in the order of A ₁ , B ₁ , B ₀ , A ₀ and B ₂ , motion information of available spatial candidates may be added to the merge list.

Derivation method of merge list in merge mode and skip mode

As described above, the maximum number of merge candidates in the merge list can be set. The set maximum number is indicated by N. The set number may be transmitted from the encoding device 100 to the decoding device 200. The slice header of the slice may include N. In other words, the maximum number of merge candidates of the merge list for the target block of the slice may be set by the slice header. For example, the value of N may be 5 by default.

The motion information (ie, the merge candidate) may be added to the merge list in the order of steps 1) to 4) below.

Step 1) Available spatial candidates among spatial candidates may be added to the merge list. Motion information of available spatial candidates may be added to the merge list in the order shown in FIG. 10. At this time, if the motion information of the available spatial candidate overlaps with other motion information already existing in the merge list, the motion information may not be added to the merge list. Checking whether it overlaps with other motion information existing in the list may be abbreviated as "redundancy check".

The added motion information may be up to N pieces.

Step 2) If the number of motion information in the merge list is smaller than N, and a temporal candidate is available, motion information of the temporal candidate may be added to the merge list. At this time, if the available motion information of the temporal candidate overlaps with other motion information already present in the merge list, the motion information may not be added to the merge list.

Step 3) If the number of motion information in the merge list is less than N and the type of the target slice is "B", the combined motion information generated by combined bi-prediction is added to the merge list. You can.

The target slice may be a slice including the target block.

The combined motion information may be a combination of L0 motion information and L1 motion information. The L0 motion information may be motion information that refers only to the reference picture list L0. The L1 motion information may be motion information that refers to only the reference picture list L1.

Within the merge list, L0 motion information may be one or more. Also, in the merge list, L1 motion information may be one or more.

The combined motion information may be one or more. In generating the combined motion information, which L0 motion information and which L1 motion information to use among the one or more L0 motion information and the one or more L1 motion information may be defined. The one or more combined motion information may be generated in a predefined order by combined bi-directional prediction using a pair of different motion information in the merge list. One of the pair of different motion information may be L0 motion information and the other may be L1 motion information.

For example, the combined motion information that is added first and foremost may be a combination of L0 motion information having a merge index of 0 and L1 motion information having a merge index of 1. If the motion information having the merge index of 0 is not L0 motion information, or the motion information having the merge index of 1 is not L1 motion information, the combined motion information may not be generated and added. Next, motion information to be added may be a combination of L0 motion information having a merge index of 1 and L1 motion information having a merge index of 0. The specific combinations below may follow other combinations in the video encoding / decoding field.

At this time, if the combined motion information overlaps with other motion information already present in the merge list, the combined motion information may not be added to the merge list.

Step 4) If the number of motion information in the merge list is smaller than N, zero vector motion information may be added to the merge list.

The zero vector motion information may be motion information in which the motion vector is a zero vector.

The zero vector motion information may be one or more. Reference picture indices of one or more zero vector motion information may be different from each other. For example, the value of the reference picture index of the first zero vector motion information may be 0. The value of the reference picture index of the second zero vector motion information may be 1.

The number of zero vector motion information may be the same as the number of reference pictures in the reference picture list.

The reference direction of the zero vector motion information may be bidirectional. Both motion vectors may be zero vectors. The number of zero vector motion information may be smaller among the number of reference pictures in the reference picture list L0 and the number of reference pictures in the reference picture list L1. Alternatively, when the number of reference pictures in the reference picture list L0 and the number of reference pictures in the reference picture list L1 are different, a unidirectional reference direction may be used for a reference picture index that can be applied to only one reference picture list.

The encoding apparatus 100 and / or the decoding apparatus 200 may sequentially add zero vector motion information to the merge list while changing the reference picture index.

When the zero vector motion information overlaps with other motion information already present in the merge list, the above zero vector motion information may not be added to the merge list.

The order of steps 1) to 4) described above is merely exemplary, and the order between the steps may be interchanged. Also, some of the steps may be omitted according to predefined conditions.

Derivation Method of Prediction Motion Vector Candidate List in AMVP Mode

The maximum number of predicted motion vector candidates in the predicted motion vector candidate list may be predefined. The predefined maximum number is indicated by N. For example, the predefined maximum number may be 2.

The motion information (that is, the predicted motion vector candidate) may be added to the predicted motion vector candidate list in the order of steps 1) to 3) below.

Step 1) Available spatial candidates among spatial candidates may be added to the prediction motion vector candidate list. The spatial candidates may include a first spatial candidate and a second spatial candidate.

The first spatial candidate may be one of A ₀ , A ₁ , scaled A ₀ and scaled A ₁ . The second spatial candidate may be one of B ₀ , B ₁ , B ₂ , scaled B ₀ , scaled B ₁ and scaled B ₂ .

The motion information of the available spatial candidates may be added to the prediction motion vector candidate list in the order of the first spatial candidate and the second spatial candidate. At this time, if the motion information of the available spatial candidate overlaps with other motion information already present in the prediction motion vector candidate list, the motion information may not be added to the prediction motion vector candidate list. That is, when the value of N is 2, if the motion information of the second spatial candidate is the same as the motion information of the first spatial candidate, the motion information of the second spatial candidate may not be added to the predicted motion vector candidate list.

The added motion information may be up to N pieces.

Step 2) If the number of motion information in the predicted motion vector candidate list is smaller than N, and a temporal candidate is available, the motion information of the temporal candidate can be added to the predicted motion vector candidate list. At this time, if the motion information of the available temporal candidate overlaps with other motion information already present in the prediction motion vector candidate list, the motion information may not be added to the prediction motion vector candidate list.

Step 3) If the number of motion information in the prediction motion vector candidate list is smaller than N, zero vector motion information may be added to the prediction motion vector candidate list.

The zero vector motion information may be one or more. Reference picture indices of one or more zero vector motion information may be different from each other.

The encoding apparatus 100 and / or the decoding apparatus 200 may sequentially add zero vector motion information to the predicted motion vector candidate list while changing the reference picture index.

When the zero vector motion information overlaps with other motion information already present in the predicted motion vector candidate list, the above zero vector motion information may not be added to the predicted motion vector candidate list.

The description of the zero vector motion information described above for the merge list can also be applied to the zero vector motion information. Duplicate description is omitted.

The order of steps 1) to 3) described above is merely exemplary, and the order between the steps may be interchanged. Also, some of the steps may be omitted according to predefined conditions.

12 illustrates a process of transform and quantization according to an example.

12, a quantized level may be generated by performing a transform and / or quantization process on a residual signal.

The residual signal may be generated as a difference between the original block and the prediction block. Here, the prediction block may be a block generated by intra prediction or inter prediction.

The residual signal may be transformed into the frequency domain through a transform process that is part of the quantization process.

The transform kernel used for the transformation may include various DCT kernels such as Discrete Cosine Transform (DCT) type 2 (DCT-II) and Discrete Sine Transform (DST) kernels. .

These transform kernels can perform a separable transform or a 2D (non-separable transform) non-separable transform on a residual signal. The separable transform may be a transform that performs a one-dimensional (1Dimensional) transform on a residual signal in each of a horizontal direction and a vertical direction.

DCT type and DST type used adaptively for 1D conversion may include DCT-V, DCT-VIII, DST-I and DST-VII in addition to DCT-II, respectively, as shown in Tables 3 and 4 below. have.

[Table 3]

[Table 4]

As shown in Table 3 and Table 4, a transform set may be used in deriving a DCT type or a DST type to be used for transformation. Each transform set may include a plurality of transform candidates. Each conversion candidate may be a DCT type or a DST type.

Table 5 below shows an example of a transform set applied in the horizontal direction and a transform set applied in the vertical direction according to the intra prediction mode.

[Table 5]

In Table 5, according to the intra prediction mode of the target block, the number of the vertical direction transform set and the number of the horizontal direction transform set applied to the horizontal direction of the residual signal are displayed.

As illustrated in Table 5, transform sets applied to the horizontal direction and the vertical direction may be predefined according to the intra prediction mode of the target block. The encoding apparatus 100 may perform transformation and inverse transformation on the residual signal by using a transformation included in a transformation set corresponding to the intra prediction mode of the target block. Also, the decoding apparatus 200 may perform an inverse transform on the residual signal using a transform included in a transform set corresponding to the intra prediction mode of the target block.

In this transform and inverse transform, the transform set applied to the residual signal may be determined as illustrated in Tables 3, 4, and 5, and may not be signaled. The transformation indication information may be signaled from the encoding device 100 to the decoding device 200. The transform indication information may be information indicating which transform candidate is used among a plurality of transform candidates included in the transform set applied to the residual signal.

For example, when the size of the target block is 64x64 or less, all three transform sets may be configured according to the intra prediction mode. The optimal transform method may be selected from all nine multiple transform methods due to a combination of three transforms in the horizontal direction and three transforms in the vertical direction. Coding efficiency may be improved by encoding and / or decoding a residual signal using the optimal conversion method.

At this time, for at least one of the vertical transform and the horizontal transform, information on which transform among transforms belonging to the transform set is used may be entropy-encoded and / or decoded. Truncated unary binarization may be used for encoding and / or decoding such information.

The method using various transforms as described above can be applied to a residual signal generated by intra prediction or inter prediction.

The transform may include at least one of a primary transform and a secondary transform. A transform coefficient may be generated by performing a first-order transform on the residual signal, and a second-order transform coefficient may be generated by performing a second-order transform on the transform coefficient.

The primary transform can be referred to as the primary transform. Also, the primary transform may be referred to as an adaptive multiple transform (AMT). AMT may mean that different transformations are applied to each of the 1D directions (ie, vertical and horizontal directions) as described above.

The secondary transform may be a transform for improving the energy concentration of the transform coefficients generated by the primary transform. The quadratic transform can be a separable transform or a non-separable transform, like the primary transform. The non-separable transform may be a non-separable secondary transform (NSST).

The primary transformation may be performed using at least one of a plurality of predefined transformation methods. For example, the predefined plurality of transform methods include a discrete cosine transform (DCT), a discrete sine transform (DST), and a Karhunen-Loeve Transform (KLT) -based transform. It can contain.

Also, the primary transform may be a transform having various types according to a kernel function defining DCT or DST.

For example, the primary transform includes transforms such as DCT-2, DCT-5, DCT-7, DST-7, DST-1, DST-8 and DCT-8 according to the transform kernel shown in Table 6 below. can do. In Table 6, various transform types and transform kernel functions for multiple transform selection (MTS) are illustrated.

MTS may mean that a combination of one or more DCT and / or DST conversion kernels is selected for the horizontal and / or vertical conversion of the residual signal.

[Table 6]

In Table 6, i and j may be integer values of 0 or more and N-1 or less.

A secondary transform may be performed on transform coefficients generated by performing the primary transform.

As in the first order transform, a transform set can also be defined in the second order transform. Methods for deriving and / or determining a transform set as described above can be applied not only to the primary transform, but also to the secondary transform.

The first order transform and the second order transform can be determined for a specified object.

For example, the first order transform and the second order transform may be applied to one or more signal components of a luma component and a chroma component. Whether to apply the first transform and / or the second transform may be determined according to at least one of coding parameters for a target block and / or a neighboring block. For example, whether to apply the first transform and / or the second transform may be determined by the size and / or shape of the target block.

In the encoding apparatus 100 and the decoding apparatus 200, transformation information indicating a transformation method used for a target may be derived by using specified information.

For example, the transform information may include an index of transforms to be used for the primary transform and / or the secondary transform. Or, the transform information may indicate that the primary transform and / or the secondary transform is not used.

For example, when the target of the primary transform and the secondary transform is a target block, the transform method (s) applied to the primary transform and / or the secondary transform indicated by the transform information is applied to the target block and / or neighboring block. It may be determined according to at least one of the coding parameters for.

Alternatively, transform information indicating a transform method for a specified target may be signaled from the encoding apparatus 100 to the decoding apparatus 200.

For example, whether a primary transform is used, an index indicating a primary transform, whether a secondary transform is used, an index indicating a secondary transform, and the like may be derived from the decoding apparatus 200 as transform information for one CU. have. Alternatively, transform information indicating whether a primary transform is used, an index indicating a primary transform, whether a secondary transform is used, and an index indicating a secondary transform may be signaled for one CU.

Quantization transform coefficients (ie, quantized levels) may be generated by performing quantization on a residual signal or a result generated by performing the first transform and / or the second transform.

13 shows diagonal scanning according to an example.

14 shows horizontal scanning according to an example.

15 illustrates vertical scanning according to an example.

The quantized transform coefficients may be scanned according to at least one of (up-right) diagonal scanning, vertical scanning, and horizontal scanning according to at least one of intra prediction mode, block size, and block shape. The block can be a conversion unit.

Each scanning can start at a specified starting point and end at a specified ending point.

For example, the quantized transform coefficients may be changed into a one-dimensional vector form by scanning coefficients of a block using diagonal scanning of FIG. 13. Alternatively, horizontal scanning of FIG. 14 or vertical scanning of FIG. 15 may be used instead of diagonal scanning depending on the size of the block and / or intra prediction mode.

The vertical scanning may be to scan two-dimensional block shape coefficients in a column direction. The horizontal scanning may be to scan two-dimensional block shape coefficients in a row direction.

In other words, depending on the size of the block and / or the inter prediction mode, which of the diagonal scanning, vertical scanning and horizontal scanning may be used.

13, 14, and 15, the quantized transform coefficients may be scanned in a diagonal direction, a horizontal direction, or a vertical direction.

The quantized transform coefficients may be expressed in block form. A block may include a plurality of sub-blocks. Each sub-block may be defined according to a minimum block size or a minimum block shape.

In scanning, the scanning order according to the type or direction of scanning may first be applied to sub-blocks. Also, a scanning order according to a scanning direction may be applied to quantized transform coefficients in a sub-block.

For example, as shown in FIGS. 13, 14, and 15, when the size of the target block is 8x8, the transform coefficients quantized by primary transform, secondary transform, and quantization for the residual signal of the target block Can be created. Thereafter, one of three scanning sequences may be applied to four 4x4 sub-blocks, and quantized transform coefficients may be scanned for each 4x4 sub-block according to the scanning sequence.

The encoding apparatus 100 may generate entropy-encoded quantized transform coefficients by performing entropy encoding on the scanned quantized transform coefficients, and may generate a bitstream including entropy-encoded quantized transform coefficients. .

The decoding apparatus 200 may extract entropy-encoded quantized transform coefficients from the bitstream, and may generate quantized transform coefficients by performing entropy decoding on the entropy-encoded quantized transform coefficients. The quantized transform coefficients may be arranged in a two-dimensional block form through inverse scanning. At this time, as a method of inverse scanning, at least one of a (upper right) diagonal scan, a vertical scan, and a horizontal scan may be performed.

In the decoding apparatus 200, inverse quantization may be performed on quantized transform coefficients. Depending on whether secondary inverse transformation is performed, secondary inverse transformation may be performed on a result generated by performing inverse quantization. In addition, depending on whether the first inverse transform is performed, the first inverse transform may be performed on the result generated by performing the second inverse transform. The reconstructed residual signal may be generated by performing the first inverse transform on the result generated by performing the second inverse transform.

For the luma component reconstructed through intra prediction or inter prediction, inverse mapping of a dynamic range may be performed before in-loop filtering.

The dynamic range can be divided into 16 equal pieces, and the mapping function for each piece can be signaled. The mapping function can be signaled at the slice level or tile group level.

An inverse mapping function for performing inverse mapping may be derived based on a mapping function.

In-loop filtering, storage of reference pictures, and motion compensation may be performed in a de-mapped region.

The prediction block generated through inter prediction may be converted into a mapped region by mapping using a mapping function, and the converted prediction block may be used to generate a reconstructed block. However, since intra prediction is performed in the mapped region, the prediction block generated by intra prediction can be used to generate a reconstructed block without mapping and / or demapping.

For example, when the target block is a residual block of a chroma component, the residual block may be converted into a reverse-mapped region by performing scaling on the chroma component of the mapped region.

Whether scaling is available may be signaled at the slice level or tile group level.

For example, scaling is applicable only when mapping to luma components is available, and division of luma components and division of chroma components follow the same tree structure.

 Scaling may be performed based on the average of the values of the samples of the luma prediction block corresponding to the chroma prediction block. In this case, when the target block uses inter prediction, the luma prediction block may mean a mapped luma prediction block.

By referring to the lookup table using the index of the piece to which the average of the values of the samples of the luma prediction block belongs, a value necessary for scaling can be derived.

By performing scaling on the residual block using the finally derived value, the residual block can be converted into a reverse-mapped region. Thereafter, for the chroma component block, reconstruction, intra prediction, inter prediction, in-loop filtering, and storage of the reference picture may be performed in the de-mapped region.

For example, information indicating whether mapping of the luma component and the chroma component and / or demapping is available may be signaled through a sequence parameter set.

The predictive block of the target block may be generated based on the block vector. The block vector may indicate displacement between a target block and a reference block. The reference block may be a block in the target image.

As described above, a prediction mode for generating a prediction block with reference to a target image may be referred to as an intra block copy (IBC) mode.

The IBC mode can be applied to a CU of a specified size. For example, the IBC mode can be applied to the MxN CU. Here, M and N may be 64 or less.

The IBC mode may include skip mode, merge mode, and AMVP mode. In the case of the skip mode or the merge mode, a merge candidate list may be configured, and one of the merge candidates of the merge candidates of the merge candidate list may be specified by signaling the merge index. The block vector of the specified merge candidate can be used as the block vector of the target block.

In the AMVP mode, a differential block vector can be signaled. Also, the prediction block vector may be derived from the left neighboring block and the top neighboring block of the target block. In addition, an index as to which neighboring block is used may be signaled.

The prediction block of the IBC mode may be included in the target CTU or the left CTU, and may be limited to blocks in the reconstructed region. For example, the value of the block vector can be limited such that the predictive block of the target block is located within a specified region. The specified area may be an area of three 64x64 blocks that are encoded and / or decoded before the 64x64 block including the target block. By limiting the value of the block vector in this way, memory consumption and device complexity due to the implementation of the IBC mode can be reduced.

16 is a structural diagram of an encoding apparatus according to an embodiment.

The encoding device 1600 may correspond to the encoding device 100 described above.

The encoding apparatus 1600 includes a processing unit 1610, a memory 1630, a user interface (UI) input device 1650, a UI output device 1660, and a storage unit that communicates with each other through the bus 1690. It may include (1640). Also, the encoding device 1600 may further include a communication unit 1620 connected to the network 1699.

The processing unit 1610 may be a semiconductor device that executes processing instructions stored in a central processing unit (CPU), memory 1630, or storage 1640. The processing unit 1610 may be at least one hardware processor.

The processor 1610 may generate and process signals, data, or information input to the encoding device 1600, output from the encoding device 1600, or used inside the encoding device 1600, and You can perform inspection, comparison, and judgment related to data or information. That is, in the embodiment, generation and processing of data or information, and inspection, comparison, and judgment related to data or information may be performed by the processing unit 1610.

The processing unit 1610 includes an inter prediction unit 110, an intra prediction unit 120, a switch 115, a subtractor 125, a conversion unit 130, a quantization unit 140, an entropy encoding unit 150, and inverse quantization. It may include a unit 160, an inverse transform unit 170, an adder 175, a filter unit 180 and a reference picture buffer 190.

Inter prediction unit 110, intra prediction unit 120, switch 115, subtractor 125, transform unit 130, quantization unit 140, entropy encoding unit 150, inverse quantization unit 160, At least some of the inverse transform unit 170, the adder 175, the filter unit 180, and the reference picture buffer 190 may be program modules, and may communicate with an external device or system. The program modules may be included in the encoding apparatus 1600 in the form of an operating system, application program modules, and other program modules.

The program modules may be physically stored on various well-known storage devices. Also, at least some of these program modules may be stored in a remote storage device capable of communicating with the encoding device 1600.

Program modules perform a function or operation according to an embodiment, or a routine, subroutine, program, object, component and data implementing an abstract data type according to an embodiment It may include a data structure, but is not limited thereto.

The program modules may be composed of instructions or codes performed by at least one processor of the encoding device 1600.

The processing unit 1610 includes an inter prediction unit 110, an intra prediction unit 120, a switch 115, a subtractor 125, a conversion unit 130, a quantization unit 140, an entropy encoding unit 150, and inverse quantization. An instruction or code of the unit 160, the inverse transform unit 170, the adder 175, the filter unit 180, and the reference picture buffer 190 may be executed.

The storage unit may represent the memory 1630 and / or the storage 1640. The memory 1630 and the storage 1640 may be various types of volatile or nonvolatile storage media. For example, the memory 1630 may include at least one of a ROM (ROM) 1163 and a RAM (RAM) 1632.

The storage unit may store data or information used for the operation of the encoding device 1600. In an embodiment, data or information possessed by the encoding device 1600 may be stored in the storage unit.

For example, the storage unit may store pictures, blocks, lists, motion information, inter prediction information, and bitstreams.

The encoding apparatus 1600 may be implemented in a computer system including a recording medium that can be read by a computer.

The recording medium may store at least one module required for the encoding apparatus 1600 to operate. The memory 1630 may store at least one module, and at least one module may be configured to be executed by the processing unit 1610.

A function related to communication of data or information of the encoding device 1600 may be performed through the communication unit 1620.

For example, the communication unit 1620 may transmit the bitstream to the decoding device 1700, which will be described later.

17 is a structural diagram of a decoding apparatus according to an embodiment.

The decoding device 1700 may correspond to the decoding device 200 described above.

The decoding apparatus 1700 includes a processing unit 1710, a memory 1730, a user interface (UI) input device 1750, a UI output device 1760, and a storage unit that communicates with each other through the bus 1790. (1740). Also, the decoding device 1700 may further include a communication unit 1720 connected to the network 1799.

The processing unit 1710 may be a central processing unit (CPU), a semiconductor device that executes processing instructions stored in the memory 1730 or the storage 1740. The processing unit 1710 may be at least one hardware processor.

The processor 1710 may input and output to the decoding device 1700, output from the decoding device 1700, or generate and process signals, data, or information used in the decoding device 1700, and You can perform inspection, comparison, and judgment related to data or information. That is, in the embodiment, generation and processing of data or information, and inspection, comparison, and judgment related to data or information may be performed by the processing unit 1710.

The processing unit 1710 includes an entropy decoding unit 210, an inverse quantization unit 220, an inverse transform unit 230, an intra prediction unit 240, an inter prediction unit 250, a switch 245, an adder 255, and a filter. The unit 260 and the reference picture buffer 270 may be included.

Entropy decoding unit 210, inverse quantization unit 220, inverse transform unit 230, intra prediction unit 240, inter prediction unit 250, switch 245, adder 255, filter unit 260 and At least some of the reference picture buffer 270 may be program modules, and may communicate with an external device or system. The program modules may be included in the decoding device 1700 in the form of an operating system, application program modules, and other program modules.

The program modules may be physically stored on various well-known storage devices. Also, at least some of the program modules may be stored in a remote storage device capable of communicating with the decoding device 1700.

Program modules perform a function or operation according to an embodiment, or a routine, subroutine, program, object, component and data implementing an abstract data type according to an embodiment It may include a data structure, but is not limited thereto.

The program modules may be composed of instructions or codes performed by at least one processor of the decoding device 1700.

The processing unit 1710 includes an entropy decoding unit 210, an inverse quantization unit 220, an inverse transform unit 230, an intra prediction unit 240, an inter prediction unit 250, a switch 245, an adder 255, and a filter. Instructions or codes of the sub-260 and the reference picture buffer 270 may be executed.

The storage unit may represent a memory 1730 and / or storage 1740. The memory 1730 and the storage 1740 may be various types of volatile or nonvolatile storage media. For example, the memory 1730 may include at least one of a ROM (ROM) 1173 and a RAM (RAM) 1732.

The storage unit may store data or information used for the operation of the decoding device 1700. In an embodiment, data or information possessed by the decoding apparatus 1700 may be stored in the storage unit.

For example, the storage unit may store pictures, blocks, lists, motion information, inter prediction information, and bitstreams.

The decoding apparatus 1700 may be implemented in a computer system including a recording medium that can be read by a computer.

The recording medium may store at least one module required for the decoding apparatus 1700 to operate. The memory 1730 may store at least one module, and at least one module may be configured to be executed by the processing unit 1710.

A function related to communication of data or information of the decoding device 1700 may be performed through the communication unit 1720.

For example, the communication unit 1720 may receive a bitstream from the encoding device 1600.

Convolution Neural Network (CNN) filter according to the properties of the transform block

Versatile Video Coding (VVC) can define a coding tree unit (CTU) and a coding unit (CU) for block-based video coding. The size of the CTU may be 128x128. The minimum size of the CU may be 4x4. The maximum size of the CU may be 64x64.

In VVC, the input target image may be divided into CTUs, and coding for the CTUs may be performed.

The CTU may be divided into CUs having a size of 64x64 again, and each CU may be divided into an optimal CU by splitting a QuadTree plus Binary tree and Triple Tree (QTBTT) again. .

In VVC, the concept of CU may include TU and PU. The TU may be a unit for frequency conversion and quantization on a block basis for a residual signal. The PU may be a unit of prediction.

Therefore, in VVC, 17 types of 4x4, 4x8, 4x16, 4x32, 8x4, 16x4, 32x4, 8x8, 8x16, 8x32, 16x8, 32x8, 16x16, 16x32, 32x16, 32x32 and 64x64 are used for the size of the TU. Can be.

In VVC reference software and the like, an inloop filter may be included to recover an error. In-loop filters are deblocking filters and sample adaptive offset (SAO)

The error may mean a difference between an input block (that is, a target block to be encoded) and a reconstructed block.

These errors may be statistically different depending on the nature of the TU. For example, attributes can be sized. Using the characteristics of this error, a plurality of different CNNs can be constructed, and the function of the existing in-loop filter is replaced by recovering the error using the selected CNN according to the attribute of the TU Can be.

For example, to reduce errors, a CNN based on a DenseNet's Dense Block may be configured according to the size of the TU. In addition, some weights may be shared between CNNs to reduce hyperparameters and complexity.

18 shows the amount of error according to the size of TU division and TU according to an example.

The picture on the left of FIG. 18 may indicate the structure of TU segmentation of an image.

The right figure of FIG. 18 may represent a mean squared error (MSE) in pixels between an original image and a reconstructed image.

In the figure on the right, white can represent a pixel with a large MSE. Black may indicate pixels with small MSE.

According to FIG. 18, the smaller the size of the TU, the larger the MSE tends to appear. This tendency may be because the reference software of VVC uses a binary cosine transform (DCT), and thus a region corresponding to texture is more probably divided in a TU partition.

Recently, many studies using deep learning have been conducted in the field of image processing such as computer vision and image compression. Deep learning may refer to a neural network having two or more hidden layers.

Deep learning can perform better than conventional methods in various image processing fields by imitating the human brain through the feature of nonlinearity using an activation function.

During such deep learning, CNN may be a method of extracting local features of an input image by using a convolution kernel for each layer, and transferring the extracted local features to the next layer. Through this extraction and delivery, CNN can show excellent performance in the field of image processing using deep learning.

19 shows a neural network according to an embodiment.

In an embodiment, the input image to the neural network includes: 1) a YUV 4: 2: 0 original image, and 2) a predictive image and (reconstructed) reconstructed image generated without going through an in-loop filter. ) May include a residual image. The reconstructed image may be an image generated by inverse transformation and inverse quantization.

The TU segmentation information may be information indicating segmentation of the original video into TU. The picture may be divided into a plurality of TUs according to TU segmentation information. Alternatively, the specified target block may be divided into a plurality of TUs according to TU partitioning information.

Using TU segmentation information, TUs of the original image may be sequentially input as CNNs as target TUs. Through this input, learning of the CNN for the target TU can be performed.

At the time when in-loop filtering is applied, information of neighboring pixels of the target TU may be used. For example, in learning, for a luma component, a pixel 2 apart from the target TU may be input together with the target TU, and used for learning. In addition, in learning, for the chroma component, pixels separated by 1 from the target TU may be input together with the target TU, and used for learning.

Here, pixels separated by 1 may represent pixels adjacent to the target block (ie, the distance from the target block is 1 pixel). Pixels separated by 2 may represent pixels having a distance from the target block of 2 pixels or less. Alternatively, a pixel separated by 2 may represent pixels adjacent to the target block and pixels adjacent to the adjacent pixels.

Or, pixels N apart from the target block, when the upper left coordinates of the target TU are (X, Y), the width of the target TU is W, and the height of the target TU is H, the x coordinate is abnormal of XN, X + Pixels that are less than or equal to N and whose y coordinate is greater than or equal to YN and less than or equal to Y + N can be represented.

In addition, adaptation may be pre-processed before the TU is input to the CNN. For example, adaptation may refer to processing by the deep learning network 2900 for preprocessing, which will be described later.

The adaptation may be a pre-processing performed to solve the imbalance of information on a part of the boundary of the block and to deal with a difference in size between YUV channels.

The image after adaptation may be converted into 18 channels of a feature map having half the horizontal size and half the vertical size compared to the original horizontal and vertical sizes of the luma component channel.

20 shows PSNR according to the size of TU according to an example.

In FIG. 20, the maximum signal-to-noise ratio (PSNR) according to the size and size of the TU is listed.

According to FIGS. 18 and 20, the smaller the size of the divided TUs, the smaller the error, and the larger the size, the larger the error may be.

According to this trend, TUs with a smaller size can be processed into a deeper network, and TUs with a larger size can be processed with a shallower network. That is, for a TU having a large size, it is possible to process through a shallower network because fewer errors are generated, which is a target of restoration.

In other words, the depth of the network to process the TU may be determined according to the size of the TU.

In addition, when separate networks are configured for the sizes of the TU, the networks may be too many and complicated, and thus the computational complexity may be increased. In response to this problem, a shared network can be constructed. Networks in a shared network may share some weight, and a TU with a large size may exit the shared network relatively faster than a TU with a small size.

Such a shared network may generate a more advantageous effect in complexity and performance than a separated network in which separate networks are configured according to the sizes of TUs.

In the shared network, the target TU can exit the main network of the shared network according to its size, and is an additional separate dense block that has undergone additional learning after exiting the main network. As it passes through the network, it may be configured in the form of a residual block. Here, the residual block may be understood as a corrected block subjected to processing by a network as compared to a reconstructed residual block in a general encoding device and a decoding device, and may be understood as a block reconstructed by the network.

The reconstructed block may be output by combining the residual block in units of pixels with the prediction block.

For learning of the shared network, 1) the original image and 2) the predicted image and the residual image of the reconstructed image without going through the existing in-loop filter may be used. Also, learning data may be used for each TU size. Here, the ratio between the sizes of training data (ie, the sizes of TUs used as training data) may correspond to or be similar to the ratios of the sizes of TUs of an image in encoding an actual video.

Also, for learning of the shared network, an Adam Optimizer using a Mean Absolute Error for the original image and the reconstructed image through the network may be used.

In sharing the weights of the networks, the weights (shared) may change as re-learning is performed according to the size of the TU. Therefore, after the specified learning, learning of the main network sharing the weight may not be performed, and only learning of the additional network, which is a separate additional dense block configured for each TU size, may be performed. Through this selective learning, the network can be further optimized.

21 is a structural diagram of an encoding apparatus according to an embodiment.

Compared to the encoding apparatus 100, the illustrated encoding apparatus 2100 may include an in-loop network 2110. Also, the filter unit 180 may be excluded from the encoding device 2100.

The above description of the encoding device 100 may be applied to the encoding device 2100. Common parts are omitted.

Characteristic operations of the encoding apparatus 2100 according to the addition of the in-loop network 2110 will be described below.

22 is a structural diagram of a decoding apparatus according to an embodiment.

Compared to the decoding device 200, the illustrated decoding device 2200 may include an in-loop network 2210. Also, in the decoding apparatus 2200, the filter unit 2210 may be excluded.

The above-described description of the decoding device 200 may be applied to the decoding device 2200. Common parts are omitted.

The characteristic operation of the decoding apparatus 2200 according to the addition of the in-loop network 2210 will be described below.

In embodiments, the neural network may be used to perform the function of an existing in-loop filter, or to replace the in-loop filter. According to this functional point of view, neural networks may be referred to as "in-loop networks".

In embodiments, the terms "CNN", "neural network", "network" and "in-loop network" may be used interchangeably and may be used interchangeably. In addition, "in-loop network" may be abbreviated to "CNN-based in-loop network for TU" or "CNN-based in-loop network according to the size of TU".

Depending on the embodiment, the in-loop network may be plural. For example, a plurality of in-loop networks may be configured for the sizes of the TU, respectively. When a plurality of in-loop networks are configured, the in-loop network performing processing for the TU may be an in-loop network selected according to the size of the TU among the configured in-loop networks.

In addition, according to an embodiment, the in-loop network includes a set of (note) networks performing processing on all TUs and (additional) networks performing processing on TUs of a specified size according to the size of TUs. can do.

23 is a flowchart of operation of an in-loop network according to an embodiment.

The following steps 2310 and 2320 may be performed by the encoding device 2100 and / or the decoding device 2200. In other words, steps 2310 and 2320 may constitute an encoding method performed by the in-loop network 2110 of the encoding apparatus 2100, and are performed by the in-loop network 2210 of the decoding apparatus 2200. The decoding method can be configured.

In the encoding apparatus 2100, steps 2310 and 2320 of the embodiment may replace the operation of the filter unit 180. Alternatively, steps 2310 and 2320 of the embodiment may constitute a part of the operation of the filter unit 180. That is, the steps 2310 and 2320 of the embodiment may replace the in-loop filter of the filter unit 180.

In the decoding apparatus 2200, steps 2310 and 2320 of the embodiment may replace the operation of the filter unit 260. Alternatively, steps 2310 and 2320 of the embodiment may constitute a part of the operation of the filter unit 260. That is, the steps 2310 and 2320 of the embodiment may replace the in-loop filter of the filter unit 260.

Prior to the execution of step 2310, a neural network may be configured.

The neural network may be an end-to-end deep learning based CNN. The neural network includes the above-described in-loop network 2210, an in-loop network 2220, an in-loop network 2510 according to the size of a TU to be described later, a deep learning network 2900 for pre-processing, a detachable network 3100 and sharing Network 3300, or may include an in-loop network 2510, a deep learning network 2900 for preprocessing, a detachable network 3100, and a shared network 3300.

The configuration of the neural network may include learning the neural network. For example, the encoding device 2100 may perform learning of a neural network. Through learning the neural network, parameters of the neural network may be configured.

The decoding device 2200 may configure a neural network. For example, the decoding apparatus 2200 may perform learning of a neural network. Alternatively, information for configuring a neural network may be signaled from the encoding device 2100 to the decoding device 2200 through a bitstream or the like. The information for the construction of the neural network may include values of parameters set through learning.

In step 2310, a block to which a neural network is applied may be generated using a neural network according to the size of the TU.

The block to which the neural network is applied can be understood as a reconstructed block for the TU. Hereinafter, a block to which a neural network is applied may be abbreviated as a reconstructed block. In other words, the block to which the neural network is adapted can be regarded as a block generated as the correction corresponding to the correction to the reconstructed block applied by the existing in-loop filter is performed on the residual block reconstructed through the neural network. According to this aspect, the neural network may be regarded as an in-loop filter for the TU for performing the function of the in-loop filter, and may be considered as an in-loop filter based on CNN according to the size of the TU.

Depending on the configuration of the in-loop network 2210 and the in-loop network 2220, the block to which the neural network is adapted may be understood as a rebuilt reconstructed residual block (by the neural network).

The size of the TU may be one of the attributes of the TU. The size of the TU may be an example of attributes of the TU. Thus, in an embodiment, the description of “size of TU” can be understood as being changed to “attribute of TU”.

In addition, the size of the TU may mean other values determined based on the size of the TU. For example, the size of the TU may be a calculated value using the width of the TU and the height of the TU. Here, the calculation may include maximum value, minimum value, average value, median value, sum, difference, quotient of product and division, remainder of division and comparison, and the like. The size of the TU may include the width of the TU and the length of the diagonal of the TU.

The TU may be an example of a target to which a neural network is applied. Therefore, in the embodiment, the description of "TU" can be understood as being changed to other target units used in video encoding / decoding such as CTU, CU, and PU.

The neural network according to the size of the TU may be part of the entire neural network. Alternatively, the neural network according to the size of the TU may be a neural network selected based on the size of the TU among a plurality of neural networks.

In step 2320, a block reconstructed in step 2310 may be used to generate a reconstructed block for the TU.

The existing in-loop filter may be replaced through a neural network, and an equal or higher compression rate may be provided compared to a conventional in-loop filter without using additional bits.

The reconstructed block for the TU may be part of the reconstructed image. In other words, the reconstructed image may be a combination of reconstructed blocks for TUs.

24 shows a result of dividing a CTU into CUs according to an example.

In FIG. 24, a CTU having a size of 128x128 may be divided into CUs through a structure of a quad tree (QT), a binary tree (BT), and a ternary tree (TT).

The target image may be divided into CTUs having a size of MxN, and encoding and / or decoding may be individually performed on the divided CTUs.

The CTU having a size of 128x128 illustrated in FIG. 24 may be divided into CUs again, and a QT structure, a BT structure, and a TT structure may be used in this partition.

The attribute of the unit may be signaled from the encoding device 2100 to the decoding device 2200 through a bitstream. For example, the attribute of the unit may include the maximum size of the CTU and the minimum size of the CU.

The attribute of the unit can be signaled as an element in high level syntax. For example, higher level syntax can include a parameter set and a header. The parameter set may include the sequence parameter set described above and the like. The header may include a tile header and a slice header. The properties of the unit may be changed according to the value signaled as such an element.

Or, the attribute of the unit may be determined based on other coding parameters. For example, an attribute of a unit may be determined according to a result of an operation using a plurality of coding parameters related to the unit.

The TU may be a basic unit for encoding and / or decoding residual blocks such as transform, quantization, inverse quantization, inverse transform, encoding for transform coefficients, and decoding for transform coefficients.

One TU may be divided into a plurality of TUs having a smaller size through division. Also, the size of the TU may be the same as the size of the CU and / or the size of the PU.

For TU, block-based sample (ie, pixel) data may be converted into data for a block-based frequency domain using transformations such as DCT-2, DCT-8, and DST-7. This transformation may be performed on blocks of various sizes generated by QT partitioning, BT partitioning, and TT partitioning.

The block of TU may mean a transform block having a size of LxK. Here, L and K may be the same.

25 illustrates input and output of an in-loop network according to the size of a TU according to an embodiment.

In the following, embodiments are described for a 4: 2: 0 color format. However, this color format is merely exemplary, and the embodiments are not limited to the 4: 2: 0 color format and may be used in other color formats.

The operation of the in-loop network 2510 may correspond to step 2310 described above. The operation of adder 2520 may correspond to step 2320 described above. Step 2310 may include processing of the in-loop network 2510.

In encoding an image, the in-loop network 2510 may correspond to the in-loop network 2110 of the encoding device 2100. The adder 2520 may correspond to the adder 175 of the encoding device 2100.

In decoding an image, the in-loop network 2510 may correspond to the in-loop network 2210 of the decoding device 2200. The adder 2520 may correspond to the adder 255 of the decoding device 2200.

The input to the in-loop network 2510 may be 1) a prediction block for the TU and 2) a reconstructed residual block for the TU.

The reconstructed residual block for the TU may be generated through inverse transform and inverse quantization according to the size of the TU.

The size of the reconstructed residual block for the TU may be the same as the size of the TU. The reconstructed residual block for the TU may be an area corresponding to the TU among the reconstructed residual blocks.

The size of the prediction block for the TU may be the same as the size of the TU. The prediction block for the TU may be a region corresponding to the TU among the prediction blocks.

The prediction block for TU may be a prediction block generated by intra prediction and / or inter prediction, or may be part of such a prediction block.

In encoding an image, a prediction block for a TU may correspond to a prediction block generated by the inter prediction unit 110 or the intra prediction unit 120. Alternatively, the prediction block for the TU may be an area corresponding to the TU among the prediction blocks generated by the inter prediction unit 110 or the intra prediction unit 120.

In decoding an image, a prediction block for a TU may correspond to a prediction block generated by the intra prediction unit 240 or the inter prediction unit 250. Alternatively, the prediction block for the TU may be a region corresponding to the TU among the prediction blocks generated by the intra prediction unit 240 or the inter prediction unit 250.

In encoding the image, the reconstructed residual block for the TU may correspond to the reconstructed residual block generated by the inverse transform unit 170. Alternatively, the reconstructed residual block for the TU may be an area corresponding to the TU among the reconstructed residual blocks generated by the inverse transform unit 170.

In decoding an image, the reconstructed residual block for the TU may correspond to the reconstructed residual block generated by the inverse transform unit 230. Can be. Alternatively, the reconstructed residual block for the TU may be an area corresponding to the TU among the reconstructed residual blocks generated by the inverse transform unit 230.

The output from the in-loop network 2510 may be a block to which the in-loop network 2510 for the TU is applied. Here, the block to which the in-loop network is applied may be understood as a corrected reconstructed residual block or a reconstructed block. Alternatively, the block to which the in-loop network 2510 is applied is regarded as a block generated as the correction corresponding to the correction to the reconstructed block applied by the existing in-loop filter is performed on the residual block reconstructed through the neural network. Can be.

The size of the block to which the in-loop network 2510 is applied may be the same as the size of the TU.

The adder 2520 may generate a reconstructed block for the TU by combining the prediction block for the TU and the block to which the in-loop network 2510 for the TU is applied.

The size of the reconstructed block for the TU may be the same as the size of the TU.

In encoding the image, the reconstructed block for the TU generated by the adder 2520 may correspond to the reconstructed block input to the reference picture buffer 190 described with reference to FIG. 1. Alternatively, the reconstructed block for the TU may be an area corresponding to the TU among the reconstructed blocks input to the reference picture buffer 190 described with reference to FIG. 1.

In decoding an image, the reconstructed block for the TU generated by the adder 2520 may correspond to the reconstructed block input to the reference picture buffer 270 described with reference to FIG. 2. Alternatively, the reconstructed block for the TU may be an area corresponding to the TU among the reconstructed blocks input to the reference picture buffer 270 described with reference to FIG. 2.

26 shows an example of setting a TU as an input of an in-loop network for a luma component when a 4: 2: 0 color format according to an example is used.

The Y block may represent the luma (Y) component of the TU. The size of the Y block may be MxN.

In processing the Y block, the prediction block for the TU and the reconstructed residual block for the TU may correspond to an area including neighboring pixels of the TU, in addition to the area of the TU.

In FIG. 26, a region corresponding to the region of the prediction block for the TU and the region of the reconstructed residual block for the TU is illustrated by a dotted line surrounding the region of the TU.

The prediction block for the TU and the reconstructed residual block for the TU may include pixels separated by n from a region corresponding to the region of the TU. For example, n can be 2. n may vary depending on the composition of the block.

For example, when the size of the Y block is MxN, the size of the prediction block for the TU and the reconstructed residual block for the TU (including padding of 2 pixels up, down, left, and right) (M + 4) x ( N + 4). When the coordinates of the upper left of the Y block are (X, Y), the coordinates of the upper left of the prediction block for TU and the coordinates of the upper left of the reconstructed residual block for TU are (X-2, Y-2) Can be

In other words, in applying the in-loop network 2510, a region corresponding to the region of neighboring pixels of the TU, in addition to the region corresponding to the region of the TU, is used as a prediction block for the TU and a reconstructed residual block for the TU, Information of the region corresponding to the region of neighboring pixels of the TU may be used together with the prediction block for the TU and the reconstructed residual block for the TU. Alternatively, an area corresponding to an area of neighboring pixels of the TU may be added to a prediction block for the TU or a reconstructed residual block for the TU through padding.

In some cases, neighboring pixels used for padding may be set to a value of zero. For example, a neighboring pixel having a value of 0 may be used when a neighboring pixel does not exist, a neighboring pixel value is not determined, or a neighboring pixel value cannot be determined.

For example, when a TU is present at a picture boundary, neighboring pixels may not exist outside the boundary. In this case, pixels inside the region of the TU can be copied or padded to neighboring pixels. The values of neighboring pixels may be obtained through such copying or padding.

The neighboring pixels can be pixels in a prediction block or pixels in a reconstructed residual block. Alternatively, by using a causality with respect to neighboring pixels, pixels that have already been reconstructed may be used as neighboring pixels. The reconstructed pixel may include a pixel adjacent to the top of the TU and a pixel adjacent to the left of the TU. Alternatively, a combination of pixels of the prediction block and pixels of the reconstructed block may be used as neighboring pixels, and a combination of pixels of the reconstructed residual block and pixels of the reconstructed block may be used as neighboring pixels.

The neighboring pixels may be excluded for blocks (that is, reconstructed blocks) to which the in-loop network generated by the in-loop network 2510 is applied. That is, neighbor pixels that are additionally used beyond the boundary of the TU may be excluded from restoration by the in-loop network 2510. In other words, the block to which the in-loop network for the Y block is applied may be a Y component block having a size of MxN.

27 shows an example of setting a TU as an input of an in-loop network for a chroma component when a 4: 2: 0 color format according to an example is used.

The Cb block and the Cr block may represent chroma Cb components and chroma Cr components of TU, respectively.

In the 4: 2: 0 color format, when the size of the Y block is MxN, the size of the Cb block (or Cr block) may be (M / 2) x (N / 2).

In processing the Cb block (or Cr block), the prediction block for the TU and the reconstructed residual block for the TU may correspond to an area including neighboring pixels of the TU, in addition to the area of the TU.

In FIG. 26, a region of a prediction block for a TU and a region corresponding to a reconstructed residual block for a TU are illustrated by dotted lines surrounding the region of the TU.

The prediction block for the TU and the reconstructed residual block for the TU may include pixels separated by n from a region corresponding to the region of the TU. For example, n may be 1.

For example, when the size of the Cb block (or Cr block) is (M / 2) x (N / 2), the size of the prediction block for the TU and the reconstructed residual block for the TU is ((M / 2) +2) x ((N / 2) +2)). When the upper left coordinates of the Cb block (or Cr block) are (X / 2, Y / 2), the upper left coordinate of the prediction block for TU and the upper left coordinate of the reconstructed residual block for TU. May be (X / 2-1, Y / 2-1).

In other words, in applying the in-loop network 2510, a region corresponding to the region of neighboring pixels of the TU, in addition to the region corresponding to the region of the TU, is used as a prediction block for the TU and a reconstructed residual block for the TU, Information of the region corresponding to the region of neighboring pixels of the TU may be used together with the prediction block for the TU and the reconstructed residual block for the TU. Alternatively, an area corresponding to an area of neighboring pixels of the TU may be added to a prediction block for the TU or a reconstructed residual block for the TU through padding.

In some cases, neighboring pixels used for padding may be set to a value of zero. For example, a neighboring pixel having a value of 0 may be used when a neighboring pixel does not exist, a neighboring pixel value is not determined, or a neighboring pixel value cannot be determined.

For example, when a TU is present at a picture boundary, neighboring pixels may not exist outside the boundary. In this case, pixels inside the region of the TU can be copied or padded to neighboring pixels. The values of neighboring pixels may be obtained through such copying or padding.

The neighboring pixels can be pixels in a prediction block or pixels in a reconstructed residual block. Alternatively, by using a causality with respect to neighboring pixels, pixels that have already been reconstructed may be used as neighboring pixels. The reconstructed pixel may include a pixel adjacent to the top of the TU and a pixel adjacent to the left of the TU. Alternatively, a combination of pixels of the prediction block and pixels of the reconstructed block may be used as neighboring pixels, and a combination of pixels of the reconstructed residual block and pixels of the reconstructed block may be used as neighboring pixels.

The neighboring pixels may be excluded for blocks (that is, reconstructed blocks) to which the in-loop network generated by the in-loop network 2510 is applied. That is, neighbor pixels that are additionally used beyond the boundary of the TU may be excluded from restoration by the in-loop network 2510. In other words, the Cb block (or a block to which the in-loop network for Cr block 0 is applied may be a Cb component block (or Cr component block) having a size of (M / 2x (N / 2)).

28 shows the size of TU according to QT segmentation, BT segmentation, and TT segmentation according to an example.

As illustrated in FIG. 28, TUs of various sizes may be generated according to QT division, BT division, and TT division.

29 illustrates a pre-processing process for input to an in-loop network according to an example.

The encoding device 2110 and the decoding device 2200 may perform pre-processing of the input in the same way for the input to the in-loop network 2510. This pre-processing can fit the sizes of the channels for a color format of YCbCr 4: 2: 0.

The operation of the deep learning network 2900 for data preprocessing may correspond to step 2310 described above, and may be performed before other operations of step 2310. Step 2310 may include processing of the deep learning network 2900 for data preprocessing.

The input to the deep learning network 2900 for data pre-processing may be a reconstructed block for the TU, a TU block undergoing inverse quantization (in other words, a reconstructed residual block for the TU) and a predictive block for the TU. Alternatively, the input to the deep learning network 2900 for data preprocessing may be an inverse quantized TU block (that is, a reconstructed residual block for the TU) and a reconstructed block for the TU.

As described above with reference to FIGS. 26 and 27, as input to the in-loop network 2510, as many as 4 pixels each of the horizontal and vertical for the Y component, and of the horizontal and vertical for the Cb component and the Cr component, respectively. Thus, as much as 2 pixels of additional information can be obtained and used.

In order to fit the output from the in-loop network 2510 to the original size of the TU, a convolution operation without padding of 2 pixels may be performed on the Y component. The convolution operation may include a RELU operation. In addition, a convolution operation without padding of 1 pixel may be performed on the Cb component and the Cr component.

Next, in order to make the size of the Y component and the size of the Cb component (or Cr component) the same (in other words, to reduce the width and height of the Y component to 1/2 each), between the feature maps for the Y component A depthwise convolution operation without considering correlations between channels may be performed.

Thereafter, L feature maps may be generated by concatenating feature maps of Y, Cb, and Cr in the channel space.

The deep learning network 2900 for data preprocessing includes a first luma convolution layer 2910, a second luma convolution layer 2920, a luma depth convolution layer 2930, and a first chroma convolution layer 2940 and channel concatenate 2950 for the luma feature map and the chroma feature map.

The deep learning network 2900 for data preprocessing may acquire additional information of pixels separated by 2 from the region of the TU for the Y component of the TU block (or the reconstructed residual block for the TU) undergoing inverse quantization. . The obtained additional information may be input to the first luma convolution layer 2910.

The deep learning network 2900 for data pre-processing additional information of pixels separated by 1 from the region of the TU for each of the Cb component and the Cr component of the dequantized TU block (that is, the reconstructed residual block for the TU). Can be obtained. Additional information on the obtained Cb component and additional information on the Cr component may be simultaneously input to the first chroma convolution layer 2940.

Each of the first luma convolution layer 2910, the second luma convolution layer 2920 and the first chroma convolution layer 2940 may be configured with a 3D convolution kernel having a size of 8 to 64 3x3xN. have. Here, N may indicate the number of channels of the input.

The input Y component may be converted into N1 first feature maps in the first luma convolution layer 2910, and N1 first feature maps may include N2 second feature maps in the second luma convolution layer 2920. Can be converted to

The N2 second feature maps may be converted into N3 third feature maps through a convolution layer 2930 for each luma depth.

Here, the convolution layer 2930 for each luma depth may be composed of a 2D convolution kernel having a size of 32 to 64 3x3.

Here, for the N2 second feature maps, the convolution layer 2930 for each luma depth may be applied while skipping by 2 pixels. In other words, in the application of the convolution layer 2930 for each luma depth, the stride may be 2. As the stride of 2 is applied, the width of the third feature map may be 1/2 of the width of the second feature map, and the height of the third feature map may be 1/2 of the height of the second feature map. Accordingly, the output from the convolution layer 2930 for each luma depth may be 1/2 for each of the width and height compared to the input to the first luma convolution layer 2910.

The input Cb component and Cr component may be converted into N4 fourth feature maps in the first chroma convolution layer 2940.

The channel chain 2950 is a channel for the output from the convolutional layer 2930 by luma depth (N3 third feature maps) and the output from the first chroma convolutional layer 2940 (N4 fourth feature maps). The concatenation of units can be performed, and the channel can be converted by applying convolution of (N3 + N4) x1x1 to the result of the concatenation. The channel chain 2950 may generate and output L feature maps through channel conversion.

30 shows a process of generating feature maps according to an example.

In Fig. 30, the symbol "[W, H, C]" denotes the width W, the height H and the number C of channels.

In FIG. 30, as an input to the deep learning network 2900 for data preprocessing, a reconstructed block for a TU having a size of 8x16 and a TU block undergoing inverse quantization having a size of 8x16 (that is, reconstruction for a TU) The residual block) has been exemplified.

Also, the number of channels output from the first luma convolution layer 2910 is 32, the number of channels output from the second luma convolution layer 2920 is 16, and the number of channels output from the first chroma convolution layer 2940 is output. The number of channels to be may be 16.

At this time, the size of the input Y component (width x height x number of channels) may be 12x20x2. The size of the input Cb component and Cr component may be 6x10x4.

Also, the size of the kernel of the first luma convolution layer 2910 (number of input channels x width x height x number of output channels) may be 2x3x3x32. The size of the feature map that has passed through the first luma convolution layer 2910 may be 10 × 18 × 32.

Also, the size of the kernel of the second luma convolution layer 2920 may be 32x3x3x16. The size of the feature map that has passed through the second luma convolution layer 2920 may be 8 × 16 × 16.

In addition, the kernel size of the convolution layer 2930 for each luma depth may be 3x3x4. The size of the feature map through the convolutional layer 2930 by luma depth may be 4x8x64 by skipping by 2 pixels (that is, the stride is 2).

Also, the size of the kernel of the first chroma convolution layer 2940 may be 4x3x3x16. The size of the feature map that has passed through the first chroma convolution layer 2940 may be 4x8x16.

The result of concatenating the result from the convolutional layer 2930 for each luma depth and the result from the first chroma convolutional layer 2940 in units of channels may be feature maps having a size of 4x8x80. .

L feature maps may be finally completed by channel compression of feature maps having a size of 4x8x80, which is a result of chaining by channel, through convolution of 80x1x1xL.

Through the same processing as described above, correlation between channels can be used in processing for the YCbCr (4: 2: 0) signal without upsampling CbCr.

31 shows an in-loop separable network according to a TU size using a pre-processing network according to an example.

The TU may go through the deep learning network 2900 for data preprocessing according to the size of the TU, and then go through the separable network 3100 again according to the size of the TU. Here, a plurality of separable networks may be configured respectively according to the sizes of TU. The TU may go through a detachable network 3100 corresponding to its size.

The separable network 3100 may be regarded as a deep learning based in-loop filter according to the size of the TU. The in-loop network 2510 can include a separable network 3100. Operation of the detachable network 3100 may correspond to step 2310 described above. Or, step 2310 may include processing of detachable network 3100.

The separable network 3100 may be composed of 2 to 16 convolutional layers.

Feature maps may be input to the separable network 3100.

Images of six channels may be output from the separable network 3100. The six channels may include a sub Y ₁ channel, a sub Y ₂ channel, a sub Y ₃ channel, a sub Y ₄ channel, a Cb channel and a Cr channel.

Hereinafter, the image may mean a block, an image representing a block, or an image corresponding to a block. That is, when the encoding apparatus 2100 and the decoding apparatus 2200 perform processing on a portion of a picture, such as a TU, a block may represent a unit of such performance, and an image may be a block input into a network or a network. Can mean the output of The output image may be regarded as a block again in the encoding device 2100 and the decoding device 2200.

Among the 6 channels, 4 channels (ie, sub Y ₁ channel, sub Y ₂ channel, sub Y ₃ channel, and sub Y ₄ channel) are the Y component (or Y component) of the reconstructed image of one channel. Image). In addition, the Cb component (or the Cb component reconstructed image) of the reconstructed image and the Cr component of the reconstructed image (or, the remaining 2 channels of the 6 channels (ie, Cb channel and Cr channel)) Cr component reconstructed image), respectively.

The deep learning network 2900 for data pre-processing may be plural, and the separable network 3100 may be plural. The deep learning networks for a plurality of data preprocessing and the plurality of separable networks may be configured separately for sizes of the TU. One of the deep learning networks for a plurality of data pre-processing and one of the plurality of separable networks may be selected according to the size of the TU.

The size of the TU can include width and height. For example, the sizes of the TU can include 4x4, 4x8, 4x16, 4x32, 8x4, 16x4, 32x4, 8x8, 8x16, 8x32, 16x8, 32x8, 16x16, 16x32, 32x16, 32x32 and 64x64.

In the configuration of multiple separable networks, the same separable network 3100 may be used for TUs having the same width and height. For example, for TUs with sizes of 4x4, 4x8 (8x4), 4x16 (16x4), 4x32 (32x4), 8x8, 8x16 (16x8), 8x32 (32x8), 16x16, 16x32 (32x16), 32x32 and 64x64 , A separable network 3100 corresponding to the size of each TU may be configured.

For example, a TU having a size of Y component of 8x16 may be converted into a feature map of 4x8x80 through a deep learning network 2900 for data preprocessing, and again, this feature map is sized to 8x16 among a plurality of separable networks. Can be input into the detachable network 3100. An image of 4x8x6 may be output through the separable network 3100 for an 8x16 size. By applying the processing described with reference to FIG. 30 to the output image, a YCbCr image of size 8x16 can be generated.

32 illustrates a method of reconstructing the output of a separable network according to an example as a reconstructed image.

As described above with reference to FIG. 31, a sub Y ₁ channel, a sub Y ₂ channel, a sub Y ₃ channel, and a sub Y ₄ channel can be output from the separable network 3100.

The Y image can be reconstructed by using the sub Y ₁ channel, the sub Y ₂ channel, the sub Y ₃ channel, and the sub Y ₄ channel.

As illustrated in FIG. 32, pixels (x, y) of the sub Y ₁ channel may be pixels (2x, 2y) of the reconstructed Y image. The pixels (x, y) of the sub Y ₂ channel may be pixels (2x + 1, 2y) of the reconstructed Y image. The pixels (x, y) of the sub Y ₃ channel may be pixels (2x, 2y + 1) of the reconstructed Y image. The pixels (x, y) of the sub Y ₄ channel may be pixels (2x + 1, 2y + 1) of the reconstructed Y image.

That is, the pixels at the same position of the sub Y ₁ channel, the sub Y ₂ channel, the sub Y ₃ channel, and the sub Y ₄ channel may constitute a 2x2 area in the Y image.

33 illustrates the operation of a shared network according to an embodiment.

In FIG. 33, processing performed according to the size of the TU in the in-loop network according to the size of the TU is described. Different processes may be performed on TUs having different sizes through the in-loop network 2510 sharing a part of parameters according to the size of the TU.

In FIG. 33, the symbol “Max (M, N) = K” indicates that the maximum value (or larger value) of the width of the TU and the height of the TU is K. For example, in an embodiment, Max (M, N) may be replaced by Min (M, N) or Sum (M, N). "Min (M, N)" may represent a minimum value (or a smaller value) of the width of the TU and the height of the TU. "Sum (M, N)" may represent the sum of the width of the TU and the height of the TU.

As described above with reference to FIG. 31, different separable networks 3100 may be configured according to the size of the TU. Alternatively, a shared network 3300 in which one network is used for all sizes of the TU may be configured.

The shared network 3330 may be considered as a deep learning based weight sharing in-loop filter. The shared network 3300 may reduce execution time and the amount of memory for all parameters compared to a deep learning-based in-loop filter according to the size of the TU (that is, the detachable network 3100).

The shared network 3300 may share parameters of some convolution layers with other convolution layers. Through this sharing, the shared network 3300 can perform processing for TUs of all sizes in one network.

The shared network 3300 may be regarded as a deep learning based in-loop filter according to the size of the TU. The in-loop network 2510 may include a shared network 3300. The operation of the shared network 3300 may correspond to step 2310 described above. Or, step 2310 may include processing of the shared network 3300.

For the feature maps that have undergone the pre-processing described above, a convolution operation in which parameters for TUs having different sizes are shared may be performed, and then, as a new convolution operation specified according to the size of the TU is applied, TU can be restored.

The parameters may include kernel parameters. Hereinafter, the parameter may mean a kernel parameter. Or, the parameter can be understood as a weight.

In FIG. 33, since the MSE of the reconstructed block is the largest when Max (M, N) = 4, the layers may be configured deepest for the case of Max (M, N) = 4.

For example, when the maximum value of the width and height of the TU is 32, for a pre-processed feature map, after the convolution operation in which the networks for TUs having different sizes share parameters are applied, the maximum value is As the convolution operation with parameters for only 32 TUs is applied, the TU can be restored.

The shared network 3300 may include a main network 3310 and an additional network 3320. The main network 3310 can share parameters regardless of the size of the TU. The additional network 3320 may configure parameters for a TU of a specified size according to the size of the TU.

Feature maps may be input to the main network 3310.

The output from the additional network 3320 may be a YCbCr image, the same as the output of the detachable network 3100 described above. The size of the YCbCr image output from the additional network 3320 may be the same as the size of the image (ie, target TU) input to the deep learning network 2900 for preprocessing. That is, when images of six channels are input to the shared network 3000, the input images may be reconstructed as YCbCr images. At this time, the width and height of the image of the six channels may be 1/2 of the width and height of the image input to the deep learning network 2900 for preprocessing. Also, the width and height of the reconstructed YCbCr image may be the same as the width and height of the image input to the deep learning network 2900 for preprocessing.

The main network 3310 sharing parameters regardless of the size of the TU may be composed of four sub-networks. Each sub-network may consist of 1 to 4 convolutional layers.

In the first sub-network, parameters may be shared for all TUs. In the second sub-network, parameters may be shared for TUs except for TUs with Max (M, N) of 64. In the third sub-network, parameters may be shared for TUs except for TBs with Max (M, N) of 32 or 64. In the fourth sub-network, parameters may be shared for TUs except for TUs with Max (M, N) of 16, 32, or 64.

The additional network 3320 in which parameters are configured according to the size of the TB may be determined according to the size of the TU. The additional networks 3320 may be plural. In Figure 33, five additional networks are shown.

Each additional network of a plurality of additional networks may consist of 1 to 4 convolutional layers. For the specified TU, an additional network 3320 that fits the size of the TU among a plurality of additional networks may be selected.

In other words, the TU can go through the main network 3310 and subnetworks of the main network that meet the conditions according to the size of the TU, and then go through the additional network 3320 that fits the size of the TU, and then have the original TU size. It can be restored as an image.

For example, if the principle size of the TU is 4x4, the TU may be converted into a feature map of 2x2x18 size through a deep learning network 2900 for pre-processing. After this feature map goes through the first sub-network, the second sub-network, the third sub-network, and the fourth sub-network of the main network 3310, an additional network 3320 is processed when Max (M, N) is 4 again. ), It can be finally restored to a 2x2x6 sized image. The reconstructed image may be converted into a 4x4 YCbCr image through the process described in FIG. 32.

For example, if the original size of the TU is 8x4, the TU may be transformed into a feature map of 4x2x18 size through a deep learning network 2900 for pre-processing. After this feature map goes through the first sub-network, the second sub-network, the third sub-network, and the fourth sub-network of the main network 3310, additional networks 3320 are processed when Max (M, N) is 8 again. ), It can be finally restored to a 4x2x6 sized image. The reconstructed image may be converted into an 8x4 YCbCr image through the process described in FIG. 32.

For example, if the original size of the TU is 8x16, the TU may be converted into a feature map of 4x8x18 size through a deep learning network 2900 for pre-processing. After the feature map passes through the first sub-network, the second sub-network, and the third sub-network of the main network 3310, it goes through an additional network 3320 to process when Max (M, N) is 16 again, finally As a result, it can be restored to a 4x8x6 image. The restored image may be converted into an 8x16 YCbCr image through the process described in FIG. 32.

For example, if the original size of the TU is 32x32, the TU may be converted into a feature map of 16x16x18 size through a deep learning network 2900 for preprocessing. After the feature map passes through the first sub-network and the second sub-network of the main network 3310, and then passes through an additional network 3320 to process when Max (M, N) is 32, finally, an image of size 16x16x6 Can be restored. The reconstructed image may be converted into a 32x32 size YCbCr image through the process described in FIG. 32.

For example, if the original size of the TU is 64x64, the TU may be converted into a feature map of 16x16x18 size through a deep learning network 2900 for preprocessing. After such a feature map passes through the first sub-network of the main network 3310, and then goes through an additional network 3320 to process when Max (M, N) is 64, it may be finally restored to an image of size 32x32x6. . The reconstructed image may be converted into a 64x64 YCbCr image through the process described in FIG. 32.

As shown in FIG. 33, the larger the size of the TU, the smaller the number of convolution parameters applied to the TU, and the smaller the size of the TU, the more the number of convolution parameters applied to the TU. It can be big.

That is, for a TU of a large size, processing corresponding to an in-loop filter may be possible even if the number of convolutional layers is small, and for a TU of a small size, processing corresponding to an in-loop filter should be increased when the number of convolution layers is increased. Can be possible.

In addition, parameter sharing may be made less for a large-size TU, and parameter sharing may be made more for a small-size TU.

Even when the shared network 3300 described above is used, detailed training according to the size of the TU according to the size of each TU (eg, max (M, N)) is possible for TUs of different sizes. Can be. In addition, in determining the size of the TU, a larger value among the width of the TU and the height of the TU may be used as a criterion for determining the size. When the size is determined, the TU is processed according to its size, and the size of the TU is determined through a separate additional dense block (for example, an additional network 3320) applied separately for each of the sizes. You can go through separate treatments that fit.

Through the embodiments as described above, an in-loop shared network or an in-loop separable network according to the size of the TU may be configured.

In the above-described embodiments, in applying the specified processing to the specified object, specified conditions may be required, and when the specified processing is described as being processed under the specified determination, the specified coding parameter It may be interpreted that the specified coding parameter can be replaced with another coding parameter, if it is determined whether the specified condition is satisfied based on the determination, or if the specified decision is made based on the specified coding parameter. That is to say, a coding parameter affecting a specified condition or a specified decision can be regarded only as an example, and it is to be understood that in addition to the specified coding parameter, a combination of one or more other coding parameters serves as the above specified coding parameter. Can be.

In the above-described embodiments, the methods are described based on a flow chart as a series of steps or units, but the present invention is not limited to the order of steps, and some steps may occur in a different order or simultaneously with other steps as described above. Can be. In addition, those of ordinary skill in the art are aware that the steps shown in the flowcharts are not exclusive, other steps may be included, or one or more steps in the flowcharts may be deleted without affecting the scope of the present invention. You will understand.

The above-described embodiments include examples of various aspects. Although not all possible combinations for indicating various aspects can be described, those skilled in the art will recognize that other combinations are possible in addition to those explicitly described. Accordingly, the present invention will be said to include all other replacements, modifications and changes that fall within the scope of the following claims.

The embodiments according to the present invention described above may be implemented in the form of program instructions that can be executed through various computer components and can be recorded in a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the computer-readable recording medium may be specially designed and configured for the present invention or may be known and usable by those skilled in the computer software field.

The computer-readable recording medium may include information used in embodiments according to the present invention. For example, a computer-readable recording medium may include a bitstream, and the bitstream may include information described in embodiments according to the present invention.

The computer readable recording medium may include a non-transitory computer-readable medium.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs, DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes produced by a compiler, but also high-level language codes executable by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.

In the above, the present invention has been described by specific matters such as specific components and limited embodiments and drawings, but this is provided only to help a more comprehensive understanding of the present invention, and the present invention is not limited to the above embodiments , Those skilled in the art to which the present invention pertains can make various modifications and variations from these descriptions.

Therefore, the spirit of the present invention is not limited to the above-described embodiment, and should not be determined, and all claims that are equally or equivalently modified as well as the claims below will fall within the scope of the spirit of the present invention. Would say

Claims (1)

Generating a block to which the neural network is applied using a neural network according to the size of the transformation unit;
Generating a reconstructed block for the conversion unit using the generated block.
Video decoding method comprising a.

Images