EP3345391A2

EP3345391A2 - Method of coding and decoding images, device for coding and decoding images and computer programmes corresponding thereto

Info

Publication number: EP3345391A2
Application number: EP16770058.2A
Authority: EP
Inventors: Pierrick Philippe
Original assignee: Orange SA
Current assignee: Orange SA
Priority date: 2015-08-31
Filing date: 2016-08-26
Publication date: 2018-07-11
Also published as: WO2017037368A3; US20190028738A1; JP2020205609A; JP7130708B2; WO2017037368A2; KR20180048719A; FR3040578A1; CN115802030A; CN107925757B; BR112018003492A2; US20200137419A1; US10523969B2; JP6951324B2; BR112018003492B1; US10911784B2; JP2018529279A; CN107925757A

Abstract

The invention relates to a method of coding at least one image (ICj) sliced into blocks, implementing, for a current block (Bi) to be coded of said image: - a prediction (C3) of the current block in accordance with a prediction mode (MPs) selected from among a plurality of predetermined prediction modes, - a calculation (C4) of a residual data block representative of a difference between a predictor block (BPopt) obtained on completion of the prediction and the current block, - an application (C6) of a transform operation to the data of said residual block, said transform operation belonging to a set of transform operations which is previously stored (C0) in association with a selected prediction mode, - a coding (C8) of the data obtained on completion of said transform operation, said method of coding being characterized in that during the storage (C0) of the set of transform operations which is associated with said selected prediction mode, the number of transform operations contained in said set is different from the number of transform operations contained in a set of transform operations which is stored in association with at least one other predetermined prediction mode of said plurality.

Description

METHOD FOR ENCODING AND DECODING IMAGES, DEVICE FOR ENCODING AND DECODING IMAGES AND PROGRAMS

CORRESPONDING COMPUTER FIELD OF THE INVENTION

The present invention relates generally to the field of image processing, and more specifically to the encoding and decoding of digital images and digital image sequences.

The encoding / decoding of digital images applies in particular to images from at least one video sequence comprising:

images coming from the same camera and succeeding each other temporally (coding / decoding of 2D type),

images from different cameras oriented according to different views (coding / decoding of 3D type),

- corresponding texture and depth components

(3D type encoding / decoding),

- etc ...

The present invention applies similarly to the coding / decoding of 2D or 3D type images.

The invention may especially, but not exclusively, apply to video coding implemented in current AVC and HEVC video encoders and their extensions (MVC, 3D-AVC, MV-HEVC, 3D-HEVC, etc.), and to corresponding decoding. Prior art

Current video encoders (MPEG, H.264, HEVC, ...) use a block representation of the video sequence. The images are cut into blocks, which can be recursively redrawn. Then each block is coded by intra-image prediction or inter-image. Thus, some images are coded by spatial prediction (Intra prediction), other images are also coded by temporal prediction (Inter prediction) with respect to one or more coded-decoded reference images, by means of a compensation in movement known by those skilled in the art.

For each block is coded a residue block, also called residue of prediction, corresponding to the original block minus a prediction. Note that in a particular case, the prediction can be omitted, the residual block then being equivalent to the original block. The residue blocks are transformed using a transform mathematical operation and then quantized using a mathematical quantization operation, for example of the scalar type. For the sake of simplification, the mathematical operation of transform will be called later 'transformed' and the mathematical operation of quantification will be called later 'quantification'.

Coefficients are obtained at the end of the quantization step. They are then browsed in a reading order that depends on the coding mode that was chosen. In the HEVC standard, for example, the reading order is dependent on the prediction made and can be done in the "horizontal", "vertical" or "diagonal" order.

At the end of the aforementioned course, a one-dimensional list of coefficients is obtained. The coefficients of this list are then encoded in bits by an entropy coding whose purpose is to code the coefficients without loss.

The bits obtained after entropy coding are written in a signal or data stream which is intended to be transmitted to the decoder.

In a manner known per se, such a signal comprises:

- the quantified coefficients contained in the aforementioned list,

information representative of the coding mode used, in particular:

• the prediction mode (Intra prediction, Inter prediction, prediction by default carrying out a prediction for which no information is transmitted to the decoder ("in English" skip "));

• information specifying the type of prediction (orientation, reference image, ...);

· The type of division of the block;

• movement information if necessary;

• etc. Once the stream has been received by the decoder, the decoding is done image by image, and for each image, block by block. For each block, the corresponding elements of the stream are read. Inverse quantization and inverse transform of block coefficients are performed to produce the decoded prediction residue. Then, the prediction of the block is calculated and the block is reconstructed by adding the prediction to the decoded prediction residue.

The conventional coding / decoding technique that has just been described certainly allows improvements in coding performance. Depending on the video context, it allows:

an improvement in the quality of the images for a given bit rate of the network used to transmit the images,

a reduction in the transmission rate of the images for a previously fixed image quality criterion.

However, such coding performances are not currently optimized and are still to be improved, in particular from the point of view of obtaining the best compromise - memory resources / coding performance.

In particular, such an optimization could concern the above-mentioned transform. It is classically a linear transform which, when applied to a residual block containing a determined number of K pixels (K> 1), makes it possible to obtain a set of K coefficients. In current video encoders / decoders (MPEG, H.264, HEVC, ...), only one transform is stored in association with a given prediction mode.

In the field of video coding, the discrete cosine transforms, DCT (abbreviation of "Discrete Cosine Transform"), or the discrete sinus transforms, DST (abbreviation of "Discrete Sine Transform"), are generally preferred, especially for the following reasons:

they are block transforms and it is thus easy to manipulate the blocks independently of one another,

they are effective for compacting information in the frequency domain, where the flow reduction operation operates.

In a conventional manner, such a transform can be of separable type or not separable.

As it is a separable type transform, in a first case, a first Al transform is applied to a residue block X of K pixels which are organized in the form of an MxN matrix, where Al is a matrix of data of size MxM and M, N are natural numbers greater than or equal to 1. At the end of the application of this first transform is obtained a first transformed block Al.x.

A transposition operation t is then applied to the transformed block Al.x. At the end of this transposition, a transposed block (Al.x) ^f is obtained.

Finally, a second transform Ac is applied to the transposed block (Al.x) ^f , where Ac is a data matrix of size NxN. At the end of the application of this second transform is obtained a second transformed block X of K = NxM pixels, such that:

X = Ac - (Al - xf

In a second case, the order of application of the Al and Ac transforms is reversed. The second transformed block X of K = NxM pixels is then written as follows:

X = Al - (Ac - x ^t y

The transformed block X obtained according to this second case is similar to the transformed block X obtained according to the first case, to a close transposition.

In the particular case where the residue block x is square, that is to say M = N, the matrices Al and Ac have the same size.

On decoding, in a manner known per se, reverse transforms of those mentioned above are applied.

Thus, if the transform has been applied according to the first case, the corresponding inverse transform makes it possible to obtain the residue block x using the following calculation:

x = Al ^{'1 ■} (Ac ^{' 1 ■} xy

Thus, if the transform has been applied according to the second case, the corresponding inverse transform makes it possible to obtain the residue block x using the following calculation: x = (Ac- ^{1 ■} (Ai ^ - xyy

AI ^"1 and Ac ^{" 1} represent the respective inverse transforms of the Al and Ac transforms. They make it possible to obtain the values of the residue block x from the values of the transformed block X. The matrices AI ^"1 and Ac ^{" 1} are commonly called inverse matrices of Al and Ac respectively, in the case where the matrices are chosen orthogonal. correspond to the matrices transposed of Al and Ac respectively.

As it is a non-separable type transform, it is written in coding as the multiplication of the residue block x, put in the form of a vector of dimension 1 xK, by a matrix A of size KxK. The transformed block X obtained after the application of this transform is then written as follows:

X = A - x

At decoding, the inverse transform consists in multiplying the transformed block X by the inverse matrix A ^"1 of A which can be the transpose of A, when A is orthogonal.This inverse transform makes it possible to obtain the following residue block x:

x = A ^{~ 1} - X

In the field of video coding, it has been proposed, in particular in the publication "Non-separable mode depend transforms for Intra coding in HEVC, Adrià Arrufat et al. VCIP 2014 "to increase the number of transforms by proposed modes of prediction. Thus, in the context of a HEVC standard spatial prediction that can be implemented according to thirty-five different spatial prediction modes, it is proposed to store 16 transforms in association with each of the prediction modes respectively. When a prediction mode is selected, a transform is selected from among the 16 transforms stored for this prediction mode, according to a predetermined coding performance criterion, such as the debit-distortion criterion well known to those skilled in the art. Thus, for each prediction considered, the transform application step is better adapted to the nature of the prediction residue signals and the coding performance is improved according to the distortion criteria for a given bit rate which are well known to the man. business.

However, with such an approach, the amount of transforms to be added in order to implement such an adaptation has an impact on the memory resources to be used both by the coder who must store the transforms for each of the predictions considered, and by the decoder which must also know the transforms to apply the transform inverse of that applied to the coding.

Thus, if one places oneself in the case of separable transforms, which is the general framework in video coding, the storage need can be estimated according to a total number NT of transforms, such that NT = n ^* N _M p where n is the number of transforms to be provisioned by prediction mode and N _M p is the number of prediction modes proposed. More particularly in the case of a HEVC encoding using, for example, transforms of size 8 × 8, the coefficients of which are stored on one byte, it is necessary to store 2 transforms (vertical and horizontal) which requires at least 2 * 8 * 8 * n * N _MP ) / 1024 = 4.375 kilobytes, where n = 1 and N _M p = 35.

This configuration is shown on the first line of the table below. For this configuration, the coding performances obtained are also represented. These performances correspond to the gain in bit rate, that is to say to the percentage of reduction of the bit rate obtained without affecting the coding performance (constant distortion). The following lines represent the evolution of the memory resources required when respectively two, four, eight, sixteen 8x8 transforms are stored in association with each of the 35 spatial prediction modes, as well as the corresponding coding performance. Number n of Total number quantity of Flow Gain

memory NT transforms

/ mode of transformations needed

prediction

1 35 4.375 kb 0.84%

2 70 8.75 kb 1, 33%

4,140 17.5 KB 1, 72%

8,280 35 KB 2.10%

16,560 70 KB 2.48%

It should be noted that according to the table above, the amount of memory increases linearly with the number of transforms to be provisioned by prediction mode and becomes significant compared to the amount of storage required for current encoders, such as for example coders. HEVC, for which the amount of memory dedicated to the storage of transforms is about 1 kb.

In addition, it is observed that the highest coding performance (2.48% throughput gain) requires a significant amount of memory of 560kb over the 1kb maximum memory required in HEVC.

Even if the bit rate gain of 2.48% proves to be interesting, a memory capacity of 70kB to store 560 transforms in order to obtain such a gain proves to be too expensive, given that the memories used for the hardware implementation of Video coding / decoding systems must be fast with respect to the amount of data processed.

Object and summary of the invention

One of the aims of the invention is to overcome disadvantages of the state of the art mentioned above. For this purpose, an object of the present invention relates to a method of coding at least one image cut into blocks, implementing, for a current block to code the image:

a prediction of the current block according to a prediction mode selected from among a plurality of predetermined prediction modes,

a calculation of a data residue block representative of a difference between a predictor block obtained at the end of the prediction and the current block,

an application of a transform operation to the data of said residue block, said transform operation belonging to a set of transform operations that is previously stored in association with a selected prediction mode,

an encoding of the data obtained as a result of said transform operation

Such a coding method is remarkable in that when storing the set of transform operations associated with the selected prediction mode, the number of transform operations contained in this set is different from the number of transform operations contained in the set. in a set of transform operations that is stored in association with at least one other predetermined prediction mode of the plurality of predetermined prediction modes.

Such an arrangement makes it possible, when there are several operations of transformation by mode of prediction:

or to significantly reduce the memory resources of the encoder for storage of the transform matrices with respect to the memory resources of the prior art encoders, without thereby deteriorating the coding performance of the current block,

- Or increase the coding performance of the current block without requiring an increase in the memory resources of the encoder for storage of the transform matrices, compared to the memory resources of the encoders of the prior art. According to a particular embodiment, for at least two sets of transform operations respectively stored in association with two prediction modes of the plurality of predetermined prediction modes, the number of transform operations in each of said two sets contains in common at least one identical transform operation.

Such an arrangement further reduces the memory resources of the encoder for storing the transform matrices.

For example, as part of the HEVC standard, at least two sets of transform operations could contain in common:

a single transform operation, for example a DCT type transform,

two transform operations, for example a DCT type transform and a DST type transform,

- etc.

According to another particular embodiment, the set of transform operations associated with the selected prediction mode is stored in association with at least one other prediction mode of the plurality of predetermined prediction modes.

Such an arrangement makes it possible to further reduce the memory resources of the encoder for storing the transform matrices.

The invention proposes different ways of determining the number of transform operations according to the prediction mode, so as to optimize the compromise "memory resources-coding performance".

According to a particular embodiment, the number of transform operations that is determined in the case of a prediction mode associated with a vertical or horizontal prediction direction is greater than the number of transform operations that is determined in the case a prediction mode associated with an oblique prediction direction.

According to another particular embodiment, the number of transform operations, which is determined in the case of a prediction mode for which the prediction is calculated by averaging more than two pixels of an edge of the current block, is greater or equal to the number of transform operations that is determined for any other prediction mode. According to yet another particular embodiment, the number of transform operations that is determined in the case of a prediction mode that has been previously selected, in the plurality of predetermined prediction modes, as the prediction mode the more likely, is greater than the number of transform operations that is determined in the case of a prediction mode that has not been previously selected as the most likely prediction mode.

According to yet another particular embodiment, the number of transform operations contained in the set of stored transform operations in association with the selected prediction mode is determined according to the amount of information representative of the prediction mode. selected.

This last provision also makes it possible to calculate simply and optimally in terms of compromise "memory resources-coding performance" the number of transform operations to be used for a given predetermined prediction mode.

The various embodiments or aforementioned embodiments can be added independently or in combination with each other, to the steps of the coding method as defined above.

The invention also relates to a device for encoding at least one image divided into blocks, comprising a processing circuit which, for a current block to be coded with the image, is arranged for:

predicting the current block according to a prediction mode selected from among a plurality of predetermined prediction modes,

calculating a data residue block representative of a difference between a predictor block obtained at the end of the prediction and the current block,

applying a transform operation to the data of the residue block, the transform operation belonging to a set of transform operations that is previously stored in association with the selected prediction mode,

coding the data obtained as a result of the transform operation. The coding device according to the invention is remarkable in that the processing circuit is arranged to store the set of transform operations associated with the selected prediction mode, the number of transform operations contained in this set being different from the number of transform operations contained in a set of transform operations that is stored in association with at least one other predetermined prediction mode of the plurality of predetermined prediction modes.

Such a coding device is particularly suitable for implementing the aforementioned coding method.

The invention also relates to a method for decoding a data signal representative of at least one image divided into blocks, implementing, for a current block to be decoded:

a determination, in the data signal:

Data representative of a current residual block associated with the current block to be decoded,

A prediction mode of the current block to be decoded, this prediction mode belonging to a plurality of predetermined prediction modes,

a prediction of the current block according to the determined prediction mode,

an application of a transform operation to data representative of the residue block, such a transform operation belonging to a set of transform operations that is previously stored in association with the determined prediction mode,

a reconstruction of the current block using a predictor block obtained at the end of the prediction and data obtained following said transform operation.

Such a decoding method is remarkable in that when storing the set of transform operations associated with the determined prediction mode, the number of transform operations contained in this set is different from the number of transform operations contained in the set. in a set of transform operations that is stored in association with at least one another predetermined prediction mode of the plurality of predetermined prediction modes.

Similar to the encoder, such an arrangement is advantageous to the decoder which must also know the transforms to apply the inverse transform of that applied to the encoding. In particular, such an arrangement makes it possible, when there are several operations of transformation by mode of prediction:

or to significantly reduce the memory resources of the decoder in order to store the transform matrices with respect to the memory resources of the decoders of the prior art, without thereby deteriorating the reconstruction quality of the current block,

or to increase the reconstruction quality of the current block without requiring an increase in the memory resources of the decoder in order to store the transform matrices with respect to the memory resources of the decoders of the prior art.

According to a particular embodiment, for at least two sets of transform operations respectively stored in association with two prediction modes of the plurality of predetermined prediction modes, the number of transform operations in each of the two sets contains in common at least one identical transform operation.

According to another particular embodiment, the set of transform operations associated with the determined prediction mode is stored in association with at least one other prediction mode of the plurality of predetermined prediction modes.

According to yet another particular embodiment, the number of transform operations which is determined in the case of a prediction mode associated with a vertical or horizontal prediction direction is greater than the number of transform operations which is determined in the case of a prediction mode associated with an oblique prediction direction.

According to yet another particular embodiment, the number of transform operations, which is determined in the case of a prediction mode for which the prediction is calculated by averaging more than two pixels of an edge of the current block, is greater than or equal to the number of transform operations that is determined for any other prediction mode.

According to yet another particular embodiment, the number of transform operations which is determined in the case where the predetermined prediction mode has been previously selected, in the plurality of predetermined prediction modes, as the most predictive mode of prediction. likely, is greater than the number of transform operations that is determined in the case where the determined prediction mode has not been previously selected as the most probable prediction mode.

According to yet another particular embodiment, the number of transform operations contained in the set of stored transform operations in association with the determined prediction mode is determined according to the amount of information representative of the prediction mode. determined.

The various embodiments or aforementioned embodiments can be added independently or in combination with each other, to the steps of the decoding method as defined above.

The invention also relates to a device for decoding a data signal representative of at least one image divided into blocks, comprising a processing circuit which, for a current block to be decoded, is arranged for:

- determine, in the data signal:

Data representing a current residual block associated with the current block to be decoded,

A mode of prediction of the current block to be decoded, this prediction mode belonging to a plurality of predetermined prediction modes,

predict the current block according to the determined prediction mode,

applying a transform operation to the data representative of the residue block, such a transform operation belonging to a set of transform operations that is previously stored in association with the determined prediction mode, reconstructing the current block using a predictor block obtained at the end of the prediction and data obtained following the transformation operation.

The decoding device according to the invention is remarkable in that the processing circuit is arranged to store the set of transform operations associated with the determined prediction mode, the number of transform operations contained in this set being different from the number of transform operations contained in a set of transform operations that is stored in association with at least one other predetermined prediction mode of the plurality of predetermined prediction modes.

Such a decoding device is particularly suitable for implementing the aforementioned decoding method.

The invention also relates to a computer program comprising instructions for implementing one of the coding and decoding methods according to the invention, when it is executed on a computer.

This program can use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other form desirable shape.

The invention also relates to a computer-readable recording medium on which a computer program is recorded, this program comprising instructions adapted to the implementation of one of the coding or decoding methods according to the invention. as described above.

The invention also relates to a computer-readable recording medium on which a computer program is recorded, this program comprising instructions adapted to the implementation of the coding or decoding method according to the invention, as described. above.

The recording medium may be any entity or device capable of storing the program. For example, the medium may include storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a USB key or a hard disk. On the other hand, the recording medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention can be downloaded in particular on an Internet type network.

Alternatively, the recording medium may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the aforementioned coding or decoding method.

Brief description of the drawings

Other features and advantages will appear on reading a preferred embodiment described with reference to the figures in which:

FIG. 1 represents the steps of the coding method according to the invention,

FIG. 2 represents an embodiment of a coding device according to the invention,

FIG. 3 represents the steps of a method for determining a variable number of transforms by prediction mode, according to one embodiment of the invention,

FIG. 4 represents a table illustrating the number of transforms determined by prediction mode, after successive addition of sixteen transforms according to the determination method of FIG. 3;

FIG. 5 represents a comparison diagram between the compromise - memory resources / coding performance - as obtained following the implementation of the iterative process of FIG. 3, and that obtained with a number of identical transforms by prediction mode,

FIG. 6A represents a first example of grouping of various Intra prediction modes according to the symmetry of their corresponding angles,

FIG. 6B represents a second example of grouping of various Intra prediction modes according to the symmetry of their corresponding angles, FIG. 7 represents a graph which illustrates the relationship between thirty-three Intra HEVC prediction modes and their corresponding angular directions,

FIG. 8 represents an example of grouping of different modes of prediction Inter according to the symmetry of their corresponding angles,

FIG. 9 represents a table illustrating the number of transforms determined by group of prediction modes, after successive addition of sixteen transforms according to a variant of the determination method of FIG. 3,

FIG. 10 represents an embodiment of a decoding device according to the invention,

- Figure 1 1 represents the main steps of the decoding method according to the invention.

Detailed description of the coding part

An embodiment of the invention will now be described, in which the coding method according to the invention is used to code an image or a sequence of images according to a bit stream close to that obtained by a conforming coding. any of the current or future video coding standards.

In this embodiment, the coding method according to the invention is for example implemented in a software or hardware way by modifications of an encoder initially conforming to any one of the current or future video coding standards. The coding method according to the invention is represented in the form of an algorithm comprising steps C0 to C9 as represented in FIG.

According to the embodiment of the invention, the coding method according to the invention is implemented in a coding device CO represented in FIG.

As illustrated in FIG. 2, such an encoder device comprises:

an input ENT_C to receive a current image to be encoded, a processing circuit CT_C for implementing the coding method according to the invention, the processing circuit CT_C containing:

A memory MEM_C including a memory buffer MT_C,

"A PROC_C processor controlled by a computer program PG_C,

an output SOR_C for delivering a coded stream containing the data obtained at the end of the coding of the current image.

At initialization, the code instructions of the computer program PG_C are for example loaded into a RAM memory, MR_C, before being executed by the processing circuit CT_C.

The coding method shown in FIG. 1 applies to any current image IC _j fixed or part of a sequence of L images IC-i,

IC _j IC _L (1≤j≤L) to be coded.

During a step C1 shown in FIG. 1, in a manner known per se, a current image IC _{j is} partitioned into a plurality of blocks Bi, B ₂ , B ,, ..., B _s (1 <i≤S), for example of size MxM pixels, where M is a natural integer greater than or equal to 1. Such a partitioning step is implemented by a partitioning software module MP_C shown in FIG. 2, which module is controlled by the processor PROC_C.

It should be noted that for the purposes of the invention, the term "block" means coding unit (coding unit). This last terminology is notably used in the standard HEVC "ISO / IEC / 23008-2 Recommendation ITU-T H.265 High Efficiency Video Coding (HEVC)".

In particular, such a coding unit groups together sets of pixels of rectangular or square shape, also called blocks, macroblocks, or sets of pixels having other geometrical shapes.

Said blocks Bi, B ₂ , B ,, ..., B _s are intended to be coded according to a predetermined order of travel, which is for example of the lexicographic type. This means that the blocks are coded one after the other, from left to right. Other types of course are of course possible. Thus, it is possible to cut the image IC _j into several subimages called slices and to independently apply a division of this type on each sub-image. It is also possible to code not a succession of lines, as explained above, but a succession of columns. It is also possible to browse the rows or columns in one direction or the other.

Each block can also be divided into sub-blocks which are themselves subdividable.

During a step C2 represented in FIG. 1, the coder CO selects as current block a first block to be coded B, of the image IC _j , such as for example the first block B-.

During a step C3 shown in FIG. 1, the current block B is predicted by known Intra and / or Inter prediction techniques. For this purpose, the block B is predicted with respect to at least one predictor block according to a prediction mode MP _S selected from among a plurality of predetermined prediction modes MP ₀ , MP-1, ..., MP _V ,. .., MP _R where 0 <v≤R + 1 and 0 <s≤R + 1.

In a manner known per se, block B is predicted with respect to a plurality of candidate predictor blocks. Each of the candidate predictor blocks is a block of pixels that has already been encoded or encoded and decoded. Such predictor blocks are stored beforehand in the buffer MT_C of the coder CO as represented in FIG. 2.

At the end of the prediction step C3, an optimal predictor block BP _op t is obtained following a putting into competition of said predetermined prediction modes, for example by minimizing a distortion rate criterion that is well known to the human being. job. The BP _op t block is considered as an approximation of the current block B ,. The information relating to this prediction is intended to be written in a signal or data stream to be transmitted to a decoder. Such information includes in particular the type of prediction (Inter or Intra), and if appropriate, the selected prediction mode MP _S , the partitioning type of the current block if the latter has been subdivided, the reference image index and the displacement vector used in the case where an Inter prediction mode has been selected. This information is compressed by the CO encoder.

During a step C4 shown in FIG. 1, the data relating to the current block B are compared with the data of the predictor block BP _op t. More precisely, during this step, the difference between the predictor block obtained BP _op t and the current block B i is conventionally calculated.

A set of data, called residual block Βη, is then obtained at the end of step C4.

The steps C3 and C4 are implemented by a predictive coding software module PRED_C shown in FIG. 2, which module is controlled by the processor PROC_C.

During a step C5 represented in FIG. 1, the coder CO of FIG. 2 proceeds to the determination of a transform of the residue block Br 1. Such a transform can be for example:

a direct transform such as, for example, a discrete cosine transform of the DCT type,

a direct transform such as, for example, a discrete sinus transform of the DST type,

-an optimized block transform distortion rate as presented in the publication "Rate-distortion optimized! transform competition for intra coding in HEVC ", Adrià Arrufat, Philippe Pierrick, Olivier Deforges, IEEE VCIP, Dec 2014,

a wavelet transform of the DWT type,

a transform including a lapped type covering

Transforms as presented in the publication "Reduction of blocking effects in image coding with a lapped orthogonal transform", H. Malvar,

- or any other type of exploitable transform.

Said transform belongs to a set of transforms which, during a prior storage step C0 shown in FIG. 1, is stored in association with the selected prediction mode MP _S , in the buffer memory MT_C of FIG. during this storage step: the predetermined prediction mode MP ₀ is stored in association with a set of transforms containing a number NB ₀ of transforms,

the predetermined prediction mode MP is stored in association with a set of transforms containing a number NB of transforms,

the predetermined prediction mode MP _V is stored in association with a set of transforms containing a number NB _V of transforms,

the predetermined prediction mode MP _R is stored in association with a set of transforms containing a number NB _R of transforms.

According to the invention, for at least two different prediction modes MP _a , MP _b belonging to the plurality of predetermined prediction modes MP ₀ , MPi, ..., MP _V , ..., MP _R , with 0 < a≤R + 1 and 0 <b≤R + 1, the number of transforms, noted NB _a , which is contained in the set of transforms associated with the prediction mode MP _a is different from the number of transforms, noted NB _b , which is contained in a set of stored transforms in association with the prediction mode MP _b .

According to a first exemplary embodiment, for at least two sets of transforms respectively stored in association with two prediction modes of said plurality MP ₀ , MP-i,. . . , MP _V , ..., MP _R of predetermined prediction modes, the number of transforms in each of said two sets contains in common at least one identical transform operation.

For example, in the case of encoding according to the HEVC standard, the number of transforms in each of said two sets may for example contain in common:

a DST type transform or a type transform

DCT

two transforms, one of the DCT type and the other of the DST type, - more than two transformed in common.

On the other hand, more than two sets of transforms may contain in common at least one identical transform operation. Still in the case of an encoding according to the HEVC standard, and more particularly in the case of an Intra prediction conforming to this standard which proposes thirty-five possible prediction directions DPI ₀ , DP, ..., DPI ₃₄ , the sets of transforms respectively associated with each of these thirty-five prediction directions may contain in common at least one identical transform operation.

According to a second exemplary embodiment represented in FIG. 2, the set of transforms associated with said selected prediction mode MP _S and containing a number NB _S of transforms is stored in association with at least one other prediction mode, denoted MP _U , of said plurality of predetermined prediction modes MP ₀ , MPi, ..., MP _V , ..., MP _R , with 0≤u≤R + 1.

Thus, as a function of the number NB _S of transforms that is contained in the set of transforms associated with the prediction mode MP _S selected in step C4, the aforementioned determination step C5 consists of:

or to read a single transform, denoted T _{s> k} , if the set of transforms which is associated with the prediction mode MP _S contains only one transform, that is to say in the case where NB _S = 1,

either to select one of several transformations, noted for example T _{s> k} *, with 0 <k ^* ≤N-1, if the set of transforms which is associated with the prediction mode MP _S contains more than one transform, that is to say in the case where NB _S > 1, such a selection being implemented according to a predetermined coding performance criterion, for example:

· By minimizing the rate / distortion criterion well known to those skilled in the art,

• or by minimizing only the flow,

• or by minimizing only the distortion,

Or by minimizing the flow-distortion / complexity compromise well known to those skilled in the art,

• or by minimizing only the efficiency,

• or by minimizing only the complexity. It is recalled that in a video coding context, the complexity is defined for example by counting the number of mathematical operations (addition, multiplication, binary shift) involved for calculating the transform of the coefficients of the residual block.

During a step C6 shown in FIG. 1, the residue block Br is transformed using the transform T _{s> k} or T _{s> k} *. Such an operation is performed by a transform software module MTR_C, as represented in FIG. 2, which module is controlled by the processor PROC_C. At the end of step C6, a transformed Bt block is obtained.

During a step C7 represented in FIG. 1, the data of the transformed block Bt are quantized according to a conventional quantization operation, such as, for example, a scalar or vector quantization. A block Bq of quantized coefficients is then obtained. Step C7 is performed by means of a quantization software module MQ_C as represented in FIG. 2, which module is controlled by the processor PROC_C.

In FIG. 1, the quantization step C7 is represented following the transforming step C6. However, step C7 can be integrated in step C6 which is then implemented with integers including the quantization factor.

In a manner known per se, during a step C8 shown in FIG. 1, the data of the block Bq 1 is encoded. Such coding is, for example, entropic coding of CABAC type ("Context Adaptive Binary Arithmetic Coder" in English) or else an entropy coding of arithmetic type or Huffman type. At the end of step C8 are obtained coded data d-ι, d ₂ ,..., D _w ,..., D _v (1 ww V V with natural integer V) associated with the current block B ,.

Step C8 is implemented by an encoding software module MC_C shown in FIG. 2, which module is controlled by the processor PROC_C.

During a step C9 shown in FIG. 1, the signal or data flow F which contains:

the coded data di, d?, ..., dw, dv obtained at the end of the aforementioned step C8, the IDX index of the transform T _{s> k} or T _Sik * which has been determined in step C5.

The IDX index can be written in the stream, for example in the form of a binary code. Such an arrangement is for example implemented if the transform T _{s> k} or T _Sik * is a transform which is common to the R + 1 sets of transforms respectively associated with the R + 1 predetermined prediction modes MP ₀ , MP-i, ..., MP _V , ..., MP _R. The IDX index will then be set to 1 or 0, for example 1. If, on the other hand, the transform T _{s> k} or T _{s> k} * is a transform which is not common to the R + 1 sets of transforms respectively associated with the R + 1 predetermined prediction modes MP ₀ , MP-i, .. ., MPv, ..., M PR, the IDX index will contain a first bit of value 0, followed by an additional codeword representative of the transform T _{s> k} or T _{s> k} * which has been selected in the number NB _S of transforms contained in the set of transforms associated with the prediction mode MP _S selected. The first bit of value 0 is for example coded using a CABAC encoder. The additional code word may be coded on a fixed length if the number of transforms NB _S is a power of 2. The additional codeword may also be coded on a variable length code if the number of transforms NB _S is a power of 2 or not.

The step C9 is implemented by a software module MCF data signal construction, as shown in Figure 2, which module is controlled by the PROC_C processor.

The data signal F is then delivered via the output SOR_C of the coder CO of FIG. 2, then transmitted by a communication network (not shown) to a remote terminal. This includes the decoder DO shown in FIG.

In a manner known per se, the data signal F furthermore comprises certain information encoded by the coder CO, such as the type of prediction (Inter or Intra) applied to the step C3, and, if appropriate, the prediction mode selected. , the index of the obtained predictor block BP _opt obtained at the end of step C3, denoted IBP _op t _, the partitioning type of the current block B, if the latter has been partitioned, the reference image index and the motion vector used in the Inter prediction mode. It is then proceeded to decode the Br residue block. A decoded residue block BDn is then obtained. The decoded block BDi is then constructed by adding to the optimal predictor block BP _opt the decoded residue block BDn.

It should be noted that the decoded block BD is the same as the decoded block obtained at the end of the decoding process of the image IC _j which will be described later in the description. The decoded block BD is thus made available for use by the coder CO of FIG. 2.

The coding steps C1 to C9 which have just been described above are then implemented for each of the blocks Bi, B ₂ , B ,,..., B _s to be encoded of the current image IC _j considered, in a predetermined order which is for example the lexicographic order.

With reference to FIGS. 3 to 8, a method of determining the number of transforms for each predetermined prediction mode, in the case of a HEVC coding of the Intra type, will now be described with reference to FIGS. .

Prior to the implementation of the steps of the coding method of FIG. 1, the following steps are carried out:

During a step ST1 shown in FIG. 3, a maximum number n _max of transforms that can be associated with a given intra prediction mode is determined.

In the embodiment shown, n _max = 16, this number may or may not include at least one HEVC transform common to each of the 35 intra ipm prediction modes of the HEVC standard.

During a step ST2 shown in FIG. 3, the number of transforms for each of the thirty-five intrapm prediction modes of the HEVC standard is initialized to zero. Each prediction mode has at this step only one common transform of the type of that of HEVC (DCT or DST) and zero additional transform. Such a step is summarized in the table TB1 of FIG. 4, whose first column lists the 35 prediction modes intra ipm ₀ to ipm ₃₄ and the first line lists the first fifteen iterations, during each of which is added a transformed. Following the first column of Table TB1 are also listed, for each of the first fifteen iterations:

the prediction mode (ipm) selected according to the best compromise - memory resources / coding performance -, following the addition of a given transform,

the global number (Nb) of transforms added since the first iteration,

the storage capacity (ROM) used at the encoder to store the entirety of the added transforms,

- the coding performance (BDRate) obtained, that is to say the percentage of flow reduction obtained relative to a HEVC encoder.

In TB1 array, initialization to zero is represented with iter = 0. During a step ST3 shown in FIG. 3, a first iteration is carried out 1, during which a first transform is added.

During a step ST4 shown in FIG. 3, for each of the thirty-five intra prediction modes, calculation of the coding performance and the storage capacity to be used in relation to the added transform are performed.

For a given intra prediction mode ipm _x , with 0 <x <34, and a given iteration it, with it> 0, the obtained coding performances, that is to say the gain in bit rate, are denoted Ri _t , _x and the corresponding storage capacity for storing the added transform is denoted Mj _tjX .

During a step ST5 shown in FIG. 3, for each of the thirty-five intra prediction modes, calculation of the ratio a_x between the obtained bit rate deviation and the added memory deviation, such as :

a_x = (Rit, x-Ro) / (Mit, xM ₀ ) with R ₀ and M ₀ respectively having the saved bit rate and the memory added compared to the HEVC standard, that is to say the iteration it = 0. (R ₀ = 0, M ₀ = 0).

In a step ST6 shown in FIG 3, it is proceeded to select the intra prediction mode IPM _x for which the A_X report presents the most favorable value, that is to say the rate decrease the more important for added memory resources as low as possible. As represented in the third column of the table TB1 of FIG. 4, for the first iteration it = 1, it is the value of the ratio a_0 which is the most favorable and which corresponds to the prediction mode intra 0. The index of this mode is indicated at the bottom of the third column of the table in association with:

the flow rate Ri saved compared with the HEVC standard,

the memory M added with respect to the HEVC standard, ie M ₌

0.1 ko.

During a step ST7 shown in FIG. 3, the CO encoder of FIG. 2 is updated with the transform added during this first iteration.

The steps ST3 to ST7 are then again applied for each iteration.

As illustrated in FIG. 4, the table TB1 shows how, over the first fifteen iterations carried out, the bit rate gain (BDRate) is improved as a function of the iterations, as the memory resources (ROM) increase.

For example, at the iteration iter = 15, as represented in the last column of the table TB1, it is the value of the ratio a_28 which is the most favorable and which corresponds to the prediction mode intra 28. The index of this mode is indicated at the bottom of the last column of the table in association with:

- the flow R15 saved compared to the HEVC standard,

the memory M ₅ added compared to the HEVC standard, ie M ₅ = 2Ko.

Moreover, at the iter iteration = 15, is represented the number of transforms that have been added by prediction mode. So :

for the intra 0 prediction mode, four transforms have been added,

for the intra prediction modes 10 and 16, two transforms were respectively added,

for the prediction modes within 1, 8, 9, 11, 14, 25, 27 and 28, a transform was respectively added. For the Iteration iter = 15, a total of sixteen transforms has therefore been added. Referring now to Figure 5, is shown:

directly in connection with the table TB1 of FIG. 4, a curve CB1 in dotted line, representative of the evolution of the coding performance with respect to the memory resources used, following the implementation of the iterative method of FIG. ,

a curve CB2 in solid line, representative of the evolution of the coding performance with respect to the memory resources used, as obtained with a coder of the state of the art which uses an identical number of transforms by prediction mode .

By comparing the two curves CB1 and CB2, it is noted that the coding performance, for a given memory footprint, is considerably improved when the iterative method for determining the number of transforms per prediction mode, as represented in FIG. is implemented. Correspondingly, the iterative process for determining the number of transforms per prediction mode, as represented in FIG. 3, makes it possible to obtain a significant reduction in the memory resources necessary for storing the determined transforms, for the coding performances concerned.

In order to appreciate the gain provided, according to the invention, in terms of storage capacity, for coding performances referred to, is presented the table below which compares the fixed and variable storage capacities, and the reduction. storage capacity obtained with the variable storage capacities used according to the invention.

Gain in Capacity Capacity Reduction

Fixed storage capacity of

(BDRate) (%) (ko) storage storage (%)

variable

(Kb)

0.8 4.4 1 .3 71

1 .3 8.8 4.1 53

1 .7 17.5 8.3 53

2.1 35 14.9 58

2.5 70 30.3 57

2.8 140 60.4 57

Thus, for a target rate gain of 2.8%, the storage capacity of prior art encoders that use a fixed number of transforms per prediction mode is 140 kb, whereas the storage capacity of the encoders according to the using a variable number of transforms per prediction mode is 60.4 kb. The reduction of the storage capacities of the encoders according to the invention with respect to the coders of the state of the art can thus be estimated at 57% for the gain in target bit rate of 2.8%.

Consequently, the coding method according to the invention advantageously makes it possible to obtain high coding performances, with a limited impact on the storage capacity of the transforms compared to that used in the coding methods of the prior art.

With reference to FIGS. 1 to 3, 6A, 6B, 7 to 9, an embodiment of the invention implemented in the case of a HEVC encoding of Intra type, and in which the set of transforms associated with an intra prediction mode selected in step C3 of FIG. 1 is previously stored in association with at least one other prediction mode of said plurality of predetermined prediction modes.

In this example, at the end of step C3 of FIG. 1, the optimal predictor block BP _op t obtained is associated with an optimal Intra prediction direction which is, for example, the direction DPI ₂ 2- According to the present embodiment, the transform (if unique) or the set of transforms associated with the Intra DPI22 prediction direction is previously associated, during the step CO of FIG. 1, with another direction. Intra prediction, provided that this other Intra prediction direction is symmetrical with respect to the Intra DPI ₂ 2 prediction direction. In this way, two-fold fewer transforms are stored at the CO encoder, and correspondingly, at the level of the decoder which will be described later in the following description, the advantage being a reduction of the memory encoder side and decoder.

According to a first variant, two Intra prediction directions with symmetries are associated with the same transform or set of transforms.

According to a second variant, two intra-angular prediction directions each having identical angle deviations with respect to a vertical (or horizontal) direction are associated with the same transform or set of transforms.

Thus, as represented in FIG. 6A, an Intra-angular prediction direction having an angle of 30 ° with respect to a horizontal axis AH is associated with an Intra-Angular prediction direction presenting symmetrically an angle of -30 ° with respect to this axis. horizontal. With reference to FIG. 7, these two directions of prediction Intra correspond respectively to directions DPI ₄ and DP ₆ .

Similarly, as shown in FIG. 6A, an Intra Angular prediction direction having an angle of -60 ° with respect to a horizontal axis is associated with an Intra Angular prediction direction having symmetrically an angle of -120 ° with respect to a vertical axis AV. With reference to FIG. 7, these two prediction directions Intra correspond respectively to the directions DPI ₂ o and DPI ₃₂ .

Still in accordance with the example which has just been described and according to yet another embodiment, the transform or the plurality of transforms associated with the Intra DPI22 prediction direction selected in step C3 is previously associated with three other Intra prediction directions, provided that these other Intra prediction directions are symmetrical with respect to the Intra DPI ₂ 2 prediction direction. In this way, four times fewer transforms are stored at the CO encoder, and correspondingly at the decoder which will be described later in the description below. , the advantage being a reduction of the memory encoder and decoder side.

According to a first variant, four Intra prediction directions with symmetries are associated with the same transform or set of transforms.

According to a second variant, four intra-angular prediction directions each having identical angles with respect to a vertical, horizontal and diagonal axis are associated with the same transform or the same set of transforms.

Thus, as shown in FIG. 6A, an Intra Angular prediction direction having an angle of 30 ° with respect to the horizontal axis AH is associated with:

at a direction of intra-angular prediction of an angle of -30 ° by symmetry with respect to this horizontal axis,

at a direction of intra-angular prediction having an angle of -60 ° by symmetry with respect to this horizontal axis then with respect to a diagonal axis AD,

at a direction of intra-angular prediction having an angle of -120 ° by symmetry with respect to the diagonal axis AD.

With reference to FIG. 7, these four Intra prediction directions correspond respectively to the directions DPI ₄ , DPI ₆ , DPI ₂ o and DPI ₃₂ .

The table below represents nine groups G ₁ to G ₉ of Intra HEVC prediction directions to which may be associated, in the buffer MT_C of the coder CO of FIG. 2, the same transform (if unique) or a set of transforms, as determined according to the invention. Group Index directions

Intra prediction

G, 2 34 18

G ₂ 3 33 19 17

G ₃ 4 32 20 16

G ₄ 5 31 21 15

G ₅ 6 30 22 14

G ₆ 7 29 23 13

G ₇ 8 28 24 12

G ₈ 9 27 25 1 1

G ₉ 10 26

The angular directions at 45 ° multiple (45 °, -45 ° and -135 °) as shown in FIG. 6B, respectively corresponding to the prediction directions DPI ₂ , DPI- | ₈ and DPI ₃₄ , as illustrated in Figure 7, are grouped by three and are associated with the same transform or the same set of transforms.

According to the obtaining of the aforementioned groups of prediction modes, with reference to FIG. 1, it may be necessary to proceed, prior to the step C5 of determining a transform, to a step C40 of at least one displacement. data of the residual block Br, as obtained at the end of step C4 of FIG. During step C40, each piece of data considered is moved inside the residue block while keeping its nearest neighbors.

The step C40 is implemented by a calculation software module CAL_C as represented in FIG. 2, which module is controlled by the processor PROC_C.

According to a first embodiment, a type of data displacement is a transposition, namely an exchange of the row and column coordinates of a data item of the current residue block.

According to a second embodiment, a type of data displacement is a mirror, namely an exchange of columns or rows of the current residue block. Each piece of data considered in the current residue block Br, is thus moved inside the residue block while keeping its nearest neighbors.

According to a third embodiment, a type of data displacement is a combination of the transposition and the mirror, that is to say: either the application of a displacement of transposition type followed by the application of a mirror-like displacement to the data of the current residue block,

or the application of a mirror-type displacement followed by the application of a displacement of transposition type to the data of the current residue block.

In addition, according to another embodiment of the invention, the type of data displacement in the residual block Βη is a function of the prediction mode MP _S selected.

Referring to Figure 8, there are eight different types of possible displacement numbered from 0 to 7 that allow a given pixel of the residue block Βη to keep its neighboring pixels. For example, for the pixel p7, the pixel is always surrounded by pixels p2, p3, p4, p6, p8, p10, p1 1 and p12. In particular, the type 0 mirror rotation which is applied to the residual block Bn, makes it possible to obtain a modified residual block Βιτιη which, in this case, is the same as the residual block Βη.

In addition, according to the invention, an Intra prediction direction is associated with one of the eight types of displacement, according to its grouping mode. For example :

the intra-DPI ₂ 2 and DPI ₆ prediction directions associated with the same transform or the same set of transforms are themselves respectively associated with the type 0 and type 5 mirror rotations as represented in FIG. 8,

the intra-DPI ₄ and DPI ₆ prediction directions associated with the same transform or the same set of transforms are themselves respectively associated with the type 0 and type 2 mirror rotations as represented in FIG. 8,

the intra-DPI ₂ o and DPI ₃₂ prediction directions associated with the same transform or the same set of transforms are themselves respectively associated with the type 0 and type 1 mirror rotations as represented in FIG. 8,

the intra-DPI ₄ , DPI ₆ , DPI ₂ o and DPI ₃₂ prediction directions associated with the same transform or the same set of transforms are they are respectively associated with mirror rotations of type 6, type 4, type 0 and type 1, as shown in FIG.

Referring now to FIG. 9, is illustrated the table TB2 which shows how, over the first fifteen iterations performed, the bit rate gain (BDRate) is improved as a function of the iterations, as the memory resources (ROM) increase, when at least two prediction modes are associated with the same set of transforms.

For example, at the iteration iter = 1 5, as shown in the last column of the table TB2, it is the value of the ratio a_28 which is the most favorable and which corresponds to the prediction mode intra 28. This mode is indicated at the bottom of the third column of the table in association with:

- the flow R- | ₅ saved compared to a coding method of the prior art, either .21%,

- the memory M- | ₅ added with respect to a coding method of the prior art, or M ₅ = 2.1 kb.

In addition, at the iter iteration I = 1 5, is represented the number of transforms that have been added by prediction mode or by groups of prediction modes. So :

for the prediction mode intra 1 and the aforementioned groups G ₃ and G ₅ of intra prediction directions, only one transform has been respectively added,

for the aforementioned groups G ₆ , G ₇ and G ₉ of intra prediction directions, two transforms were respectively added,

for the prediction mode intra 0 and the aforementioned group G ₈ of intra prediction directions, four transforms have respectively been added.

For the Iteration iter = 15, a total of seventeen transformations was therefore added.

Table TB2 shows a significant reduction in memory resources dedicated to storing transforms of the order of 20% compared to the coding method that does not use such a grouping of prediction modes. In order to appreciate the gain provided with such a grouping of prediction modes according to the invention, the gain being evaluated in terms of storage capacity for targeted coding performances, is presented the table below which gives a comparative of the capacities fixed storage units present in a coder of the prior art using the same number of transforms by prediction mode and variable storage capacities present in the encoder according to the invention, and the reduction of storage capacity obtained with the capacity of the variable storage.

Thus, for a gain in target rate of 1 .7%, the storage capacity of the encoders that use a fixed number of transforms per prediction mode is 17.5 kb, whereas the storage capacity of the encoders according to the invention that use a variable number of transforms per prediction mode or set of prediction modes is only 5.3 kb. The reduction of the storage capacities of the encoders according to the invention with respect to the coders of the state of the art can thus be estimated at 70% for the target rate gain of 1.7%.

According to another embodiment of the invention, the iterative determination method of the number of transforms by prediction mode or by group of prediction modes can be replaced by a method of automatically determining said number of transforms.

The automatic determination method may be necessary when, in certain coding contexts, the CO encoder of FIG. 2 is forced to operate at a given point of complexity. It is then necessary, during the coding, to reduce the number of transforms in competition to limit the choice investigations by the coder of the optimal transform, during step C5 of FIG.

According to a first exemplary embodiment, the number of transforms which is determined in the case of a prediction mode associated with a vertical or horizontal prediction direction is made greater than the number of transforms which is determined in the case of a prediction associated with an oblique prediction direction.

In the case, for example, of the HEVC standard, the person skilled in the art knows that among the thirty-five intra prediction modes available, the modes 0 (Planar) and 1 (DC) are the most smoothed modes because the prediction according to these two modes is calculated by averaging more than two pixels from an edge of the current block. As a result, the Planar and DC modes are associated, during the step C0 of FIG. 1, with a greater number of transforms than the other intra-HEVC prediction modes.

In addition, in a manner known per se, the modes 10 (Horizontal) and 26 (Vertical) are used to predict patterns of the image which are horizontal or vertical. Since such patterns are frequently found in nature (eg vertical trees, poles, horizon lines, etc.), they must be associated with a higher number of transforms than the number of transforms associated with the intra-prediction modes. obliques.

According to a second exemplary embodiment, the number of transforms which is determined in the case of a prediction mode which has been previously selected, in the plurality of predetermined prediction modes MP ₀ , MP-i,. . . , MPv, ..., M PR, as the most probable prediction mode, is greater than the number of transforms that is determined in the case of a prediction mode that has not been previously selected as a mode prediction most likely.

In the case, for example, of the HEVC standard, a list of the most probable prediction modes, called MPMs (of the "Most Probable Modes"), is established prior to coding. In particular, the modes 0 (Planar), 1 (DC) and 26 (vertical) are the prediction modes assigned by default when establishing the list. According to the invention, these three modes of intra prediction are therefore previously associated, during the storage step C0 of FIG. 1, with a greater number of transforms than the number of transforms associated with the other intra-HEVC prediction modes.

According to a third exemplary embodiment, the number of transforms contained in the set of transforms stored in association with each of the predetermined prediction modes MP ₀ , MP-i,. . . , MP _V , ..., MP _R is determined according to the amount of information representative of each of these prediction modes.

In the case, for example, of the HEVC standard, the most probable intra prediction modes such as 0 (Planar), 1 (DC) and 26 (vertical) will be reported on less bits than the other modes. According to the invention, it is therefore appropriate to associate them with a higher number of transforms than the other intra prediction modes, because these three modes will be chosen more frequently than the others and consequently, there will be a greater variety of predictor blocks to test with these three modes.

Thus, according to the invention, if the signaling of a prediction mode is carried on fewer bits, then the number of transforms, which is stored in association with this prediction mode during step C0 of FIG. , is larger than the average of the transforms over all thirty-five intra prediction modes.

Detailed description of the decoding part

An embodiment of the invention will now be described, in which the decoding method according to the invention is used to decode a signal or data stream representative of an image or a sequence of images which is suitable for be decoded by a decoder according to any of the current or future video decoding standards.

In this embodiment, the decoding method according to the invention is for example implemented in a software or hardware way by modifications of such a decoder.

The decoding method according to the invention is represented in the form of an algorithm comprising steps D0 to D9 as represented in FIG. According to this embodiment, the decoding method according to the invention is implemented in a decoding device or decoder DO represented in FIG.

As illustrated in FIG. 10, such a decoder device comprises:

an input ENT_D for receiving the data signal or current flow F to be decoded,

a CT_D processing circuit for implementing the decoding method according to the invention, the CT_D processing circuit containing:

A MEM_D memory including a buffer MT_D,

A PROC_D processor controlled by a computer program PG_D,

an output SOR_D for delivering a reconstructed current image containing the data obtained at the end of the decoding according to the method of the invention.

At initialization, the code instructions of the computer program PG_D are for example loaded into a RAM memory, MR_D, before being executed by the processing circuit CT_D.

The decoding method represented in FIG. 11 applies to a signal or data flow F representative of a current image ICj to be decoded which is fixed or which belongs to a sequence of images to be decoded.

For this purpose, information representative of the current image ICj to be decoded is identified in the data signal F received at the input ENT_D of the decoder DO and as delivered at the end of the coding method of FIG.

Referring to Figure 1 1, during a step D1, it is carried out in the determination signal F residues encoded blocks associated with each of the blocks Bi, B _2, Bi, ..., B _s previously coded according to the above-mentioned lexicographic order.

Such a determination step D1 is implemented by a flow analysis software identification module MI_D, as shown in FIG. 10, which module is controlled by the PROC_D processor.

Other types of course than the one mentioned above are of course possible and depend on the order of course chosen coding. In the example shown, the blocks Bi, B ₂ , B ,,..., B _s to be decoded have, for example, a square shape and are for example of size MxM pixels where M is a natural integer greater than or equal to 1.

Each block to be decoded can also be divided into sub-blocks which are themselves subdividable.

During a step D2 represented in FIG. 11, the decoder DO of FIG. 10 selects, as current block B, to be decoded, the first block which has been coded at the end of the coding method of FIG. .

During a step D3 represented in FIG. 11, a determination is made, for example by decoding, of the residual data di, d ₂ , ..., d _w , ..., d _v associated with the block current B, to be decoded, which were coded in step C8 of FIG. At the end of such a determination are obtained:

or a set of digital information associated with the quantized block Bq, obtained at the end of step C7,

or a set of digital information associated with the quantized modified block Bmq, obtained at the end of step C7, in the case where the step C40 for moving data of the residual block Br has been implemented at the coding .

Also during step D3, information relating to the prediction type of the current block B, as implemented during coding in step C3 of FIG. 1, which has been written in the signal, is determined. F.

For this purpose, during step D3, are determined:

the prediction mode MP _S selected in step C3 of the figure

1,

the index of the predictor block BP _op t, denoted IBP _op t, and the type of partitioning of the current block B, if the latter has been partitioned.

Such a decoding step D3 is implemented by a decoding module MD_D shown in FIG. 10, which module is controlled by the processor PROC_D.

During a step D4 represented in FIG. 11, the predictive decoding of the current block to be decoded is carried out using the IBP index _opt of the block predictor that was decoded during the aforementioned step D3. For this purpose, in a manner known per se, the selection, in association with the index IBP _op t, of selecting, in the buffer memory MT_D of the decoder DO of FIG. 10, of the corresponding BP _opt prediction block, which FIG. from a plurality of candidate predictor blocks previously stored in the MT_D buffer. Each of the candidate predictor blocks is a block of pixels that has already been decoded.

Step D4 is implemented by an inverse prediction software module PRED ^"1 _D, as shown in FIG. 10, which is controlled by the PROC_D processor.

During a step D5 represented in FIG. 11, which is implemented in the case where it is the set of data associated with the block Bq, or Bmq, coded which was obtained at the end of the In the above-mentioned step D3, dequantization of this set of data is carried out according to a conventional dequantization operation which is the inverse operation of the quantization implemented during the quantization step C7 of FIG. A set of current dequantized coefficients BDq, or a set of dequantized modified current coefficients BDmq, is then obtained at the end of step D5. Such a dequantization step is for example of scalar or vector type.

Step D5 is performed by means of an inverse quantization software module MQ ^"1 _D, as shown in FIG. 10, which module is controlled by the PROC_D processor.

During a step D6 represented in FIG. 11, the decoder DO of FIG. 10 carries out the determination of a transform of the current dequantized coefficient set BDq, or the set of dequantized modified current coefficients BD m , as obtained in step D5 above. In a manner known per se, such a transform is an inverse transform from that determined at the end of step C5 of FIG. 1, such as, for example:

a direct transform such as, for example, a discrete sinus transform of the DST type, -an optimized block transform distortion rate as presented in the publication "Rate-distortion optimized! transform competition for intra coding in HEVC ", Adrià Arrufat, Philippe Pierrick, Olivier Deforges, IEEE VCIP, Dec 2014,

a wavelet transform of the DWT type,

a transform including a Lapped Transforms type recovery as presented in the publication "Reduction of blocking effects in image coding with a lapped orthogonal transform", H. Malvar,

- or any other type of exploitable transform.

Said transform belongs to a set of transforms which, during a prior storage step D0 represented in FIG. 11, is stored in association with the prediction mode MP _S , in the buffer memory MT_D of FIG. during this storage step:

the predetermined prediction mode MP ₀ is stored in association with a set of transforms containing a number NB ₀ of transforms,

the predetermined prediction mode MPi is stored in association with a set of transforms containing a number NB of transforms,

the predetermined prediction mode MP _V is stored in association with a set of transforms containing a number NB _V of transforms; the predetermined prediction mode MP _R is stored in association with a set of transforms containing a number NB _R of transforms.

According to the invention, for at least two different prediction modes MP _a , MP _b belonging to the plurality of predetermined prediction modes MP ₀ , MPi, ..., MP _V , ..., MP _R , with 0 < a≤R + 1 and 0 <b≤R + 1, the number of transforms, noted NB _a , which is contained in the set of transforms associated with the prediction mode MP _a is different from the number of transforms transformed, noted NB _b , which is contained in a set of transforms stored in association with the prediction mode MP _b .

According to a first exemplary embodiment, for at least two sets of transforms respectively stored in association with two prediction modes of said plurality MP ₀ , MP-i,. . . , MP _V , ..., MP _R of predetermined prediction modes, the number of transforms in each of said two sets contains in common at least one identical transform.

For example, in the case of a decoding according to the HEVC standard, the number of transforms in each of said two sets may for example contain in common:

an inverse transform of a DST type transform, or a reverse transform of a DCT type transform,

an inverse transform of a DST type transform and an inverse transform of a DCT type transform,

- more than two transformed in common.

On the other hand, more than two sets of transforms may contain in common at least one identical transform operation. Still in the case of a decoding according to the HEVC standard, and more particularly in the case of an inverse inverse prediction conforming to this standard which proposes thirty-five possible prediction directions DPI ₀ , DPI-i,. . . , DPI ₃₄ , the sets of transforms respectively associated with each of these thirty-five prediction directions may contain in common at least one identical transform operation.

According to a second exemplary embodiment shown in FIG. 10, the set of transforms, associated with said prediction mode MP _S determined in the aforementioned step D3 and containing a number NB _S of transforms, is stored in the buffer memory MT_D of the FIG. 10, in association with at least one other prediction mode, denoted MP _U , of said plurality of predetermined prediction modes MP ₀ , MP 1 ..., MP _V , ..., MP _R , with 0≤u≤ R + 1.

More particularly, the above-mentioned determination step D6 consists in reading in the data signal F the IDX index of the transform T _{s> k} (NB _S = 1) or T _{s> k} * (NB _S > 1) which has was selected at the coding at the end of the above-mentioned step C5 (Figure 1), which index was written in the data signal F during the aforementioned step C8 (Figure 1).

During a step D7 represented in FIG. 11, the inverse transform of the transform T _{s> k} or T _{s> k} * is applied to the set of dequantized coefficients current BDq, or to the set of current dequantized coefficients BD mq ,, as obtained in step D5 above. At the end of step D7, a current decoded residue block BDr ,, or a decoded current residual block BDmr, is obtained.

Such an operation is performed by an inverse transform software module MTR ^"1 _D, as shown in FIG. 10, which module is controlled by the processor PROC_D.

In FIG. 11, the dequantization step D5 is represented before the inverse transformation application step D7. However, step D5 can be integrated in step D7 which is then implemented with integers including the dequantization factor.

During a step D8 represented in FIG. 11, the current block B is reconstructed, by adding to the decoded residue block BDn, obtained at the end of the aforementioned step D7, the predictor block BP _op. t that was obtained at the end of the aforementioned step D4. At the end of step D8, a current decoded block BD is obtained.

Step D8 is implemented by a software module CAL1_D shown in FIG. 10, which module is controlled by the processor PROC_D.

During a step D9 represented in FIG. 11, said current decoded block BD is written in a decoded image ID _j .

Such a step is implemented by an image reconstruction URI software module as shown in FIG. 10, said module being controlled by the PROC_D processor.

The decoding steps which have just been described above are implemented for all the blocks Bi, B ₂ , B ,,..., B _s to be decoded from the current image IC _j considered, in a predetermined order which is for example the lexicographic order. In the same way as in coding, with reference to FIGS. 3 to 8, the decoding method implements, according to a first embodiment of the invention, a method for determining the number of transforms for each predetermined prediction mode. , in the case of an Intra type HEVC decoding.

With reference to FIGS. 3, 6A, 6B, 7 to 11, an embodiment of the invention implemented in the case of Intra-type HEVC decoding, and in which the set of FIG. Transforms associated with an intra prediction mode determined in step D3 of FIG. 11 are previously stored in association with at least one other prediction mode of said plurality of predetermined prediction modes.

In this example, at the end of step D3 of FIG. 11, the optimal predictor block BP _op t obtained is associated with an optimal Intra prediction direction which is, for example, the direction DPI _22. realization, the transform (if unique) or the set of transforms associated with the direction of prediction Intra DPI ₂₂ is previously associated, during the step DO of Figure 1 1, to another direction of prediction Intra , provided that this other Intra prediction direction is symmetrical with respect to the Intra DPI prediction direction ₂₂ - In this way, two times fewer transforms are stored at the decoder DO, the advantage being a reduction of the memory decoder side.

According to a second variant, two intra-angular prediction directions each having identical angles with respect to a vertical (or horizontal) direction are associated with the same transform or set of transforms.

Thus, as represented in FIG. 6A, an Intra-angular prediction direction having an angle of 30 ° with respect to a horizontal axis AH is associated with an Intra-Angular prediction direction presenting symmetrically an angle of -30 ° with respect to this axis. horizontal. In reference in FIG. 7, these two prediction directions Intra correspond respectively to directions DPI ₄ and DP ₆ .

Still in accordance with the example which has just been described and according to yet another embodiment, the transform or the plurality of transforms associated with the direction of Intra DPI prediction ₂ 2 determined in step D3 is previously associated with three other prediction directions

Intra, provided that these other prediction directions Intra are symmetrical with respect to the Intra DPI ₂ 2 prediction direction. In this way, four times fewer transforms are stored at the decoder DO, the advantage being a reduction of the decoder side memory.

at a direction of intra-angular prediction having an angle of -120 ° by symmetry with respect to the diagonal axis AD. With reference to FIG. 7, these four prediction directions Intra correspond respectively to the directions DPI ₄ , DPI- | ₆ , DPI ₂ o and DPI ₃ 2-

The table below represents nine groups G ₁ to G ₉ of Intra HEVC prediction directions to which may be associated a single transform (if unique) or a set of transforms, as determined according to the invention.

45 ° multiple angular directions (45 °, -45 ° and -135 °) as shown in FIG. 6B, corresponding respectively to the prediction directions DPI ₂ , DPI ₈ and DPI ₃₄ , as illustrated in FIG. 7, are grouped by three and are associated, in the buffer MT_D of the decoder DO of Figure 10, the same transform or the same set of transforms.

According to the obtaining of the aforementioned groups of prediction modes, with reference to FIG. 11, in the case where it is the current decoded modified residue block BDmn which was obtained at the end of the step D7, it this step is followed by a step D70 of at least one displacement of the data of the current decoded modified residual block BDmn, each considered datum being moved inside this block while keeping its nearest neighbors.

The step D70 is implemented by a calculation software module CAL2_D as represented in FIG. 10, which module is controlled by the processor PROC_D.

According to a first embodiment, a type of data displacement is a transposition, the reverse of that performed at the coding, namely an exchange of the row and column coordinates of a data block residue modified decoded BDmn current.

According to a second embodiment, a type of data displacement is a mirror, the reverse of that carried out at the coding, namely an exchange of the columns or lines of the current decoded modified residual block BDmn. Each data considered in the current decoded modified residue block BDmn is thus moved inside the latter, while keeping its nearest neighbors.

According to a third embodiment, a type of data displacement is a combination of the inverse transposition and the inverse mirror, that is to say:

or the application of an inverse transposition type displacement followed by the application of an inverse mirror displacement to the data of the current decoded modified residual block BDmn,

or the application of an inverse mirror displacement followed by the application of an inverse transposition displacement to the data of the current decoded modified residual block BDmn.

In addition, according to another embodiment of the invention, the type of data displacement in the current decoded modified residual block BDmn is a function of the prediction mode MP _S determined in the aforementioned step D3.

In a manner corresponding to the different types of displacement implemented in the coding, as represented in FIG. 8, there are also at decoding eight different types of possible displacement numbered from 0 to 7 which make it possible for a given pixel of the current decoded modified residual block. BDmn to keep its neighboring pixels. For example, for the pixel p7, the pixel is always surrounded by pixels p2, p3, p4, p6, p8, p10, p1 1 and p12. In particular, the inverse mirror rotation of the type 0 mirror rotation does not cause a data shift since the data of the current decoded modified residual block BDmn are arranged in the same order as the data of the current residue block Βη.

In addition, according to the invention, an Intra prediction direction determined at the end of the step D3 of FIG. 11 is associated with one of the eight types of inverse displacement, according to the group to which it belongs. For example : the directions of intra prediction DPI22 and DPI ₆ associated with the same transform or the same set of transforms are themselves respectively associated with the inverse mirror rotations respectively of the type 0 and type 5 mirror rotations as represented in FIG. 8,

the intra-DPI ₄ and DP ₆ prediction directions associated with the same transform or the same set of transforms are themselves respectively associated with the inverse mirror rotations respectively of the type 0 and type 2 mirror rotations as represented in FIG. 8 ,

the directions of intra-DPI ₂ o and DPI 32 prediction associated with the same transform or the same set of transforms are themselves respectively associated with the inverse mirror rotations respectively of the type 0 and type 1 mirror rotations as represented in FIG. 8 ,

the directions of prediction within DPI ₄ , DPI- | ₆ , DPI ₂ o and DPI32 associated with the same transform or the same set of transforms are themselves associated respectively with the inverse mirror rotations respectively of the type 6, type 4, type 0 and type 1 mirror rotations, such as represented in FIG. 8.

According to another embodiment of the invention, in the same way as at coding, the iterative determination method of the number of transforms by prediction mode or by group of prediction modes can be replaced by a method of automatic determination of said number of transforms.

The automatic determination method may be necessary when, in certain decoding contexts, the decoder DO of FIG. 10 is constrained to operate at a given point of complexity. This given point of complexity can be indicated by the encoder CO of FIG. 2, by a given configuration information for a portion of an image, an entire image or for a sequence of images. It can comprise a maximum number of transforms identifier or a ratio of the number of transforms relative to the number of transformations available to the decoder. Thus, during the step D6 of FIG. 11, the decoder DO searches in the buffer memory MT_D of FIG. 10 only in the number of transforms associated with FIG. said configuration information signaled, the transform selected at the coding and corresponding to the IDX index determined in step D3.

In the case, for example, of the HEVC standard, the person skilled in the art knows that among the thirty-five intra prediction modes available, the modes 0 (Planar) and 1 (DC) are the most smoothed modes because the prediction according to these two modes is calculated by averaging more pixels than the edges of the current block. Consequently, the Planar and DC modes are associated, during the step D0 of FIG. 11, with a greater number of transforms than the other intra-HEVC prediction modes.

Moreover, in a manner known per se, modes 10 and 26 are used to predict image patterns that are horizontal or vertical. Since such patterns are frequently found in nature (eg vertical trees, poles, horizon lines, etc.), they must be associated with a higher number of transforms than the number of transforms associated with the intra-prediction modes. obliques.

According to a second exemplary embodiment, the number of transforms that is determined in the case of a prediction mode that has been previously selected, in the plurality of predetermined prediction modes MP ₀ , MP-i, ..., MPv, ..., M PR, as the most probable prediction mode, is greater than the number of transforms that is determined in the case of a prediction mode that has not been previously selected as a prediction mode. more likely.

In the case, for example, of the HEVC standard, a list of the most probable prediction modes, called MPMs (of the "Most Probable Modes"), is established prior to coding. In particular, the modes 0 (Planar), 1 (DC) and 26 (vertical) are the prediction modes assigned by default when establishing the list. According to the invention, these three intra prediction modes are therefore previously associated, during the step of storage OD of Figure 1A, to a number of transforms higher than the number of transforms associated with the other modes of intra-HEVC prediction.

Thus, according to the invention, if the signaling of a prediction mode is carried on fewer bits, then the number of transforms, which is stored in association with this prediction mode during the step D0 of FIG. 1, is larger than the average of the transforms on all thirty-five intra prediction modes.

It goes without saying that the embodiments which have been described above have been given for purely indicative and non-limiting purposes, and that many modifications can easily be made by those skilled in the art without departing from the scope. of the invention

Claims

1. Method for encoding at least one image (ICj) cut into blocks, implementing, for a current block (B,) to be encoded of said image:

a prediction (C3) of the current block according to a prediction mode (MP _S ) selected from among a plurality of predetermined prediction modes,

a calculation (C4) of a data residue block representative of a difference between a predictor block (BP _opt ) obtained at the end of the prediction and the current block,

an application (C6) of a transform operation to the data of said residue block, said transform operation belonging to a set of transform operations that is previously stored (C0) in association with a selected prediction mode,

a coding (C8) of the data obtained as a result of said transform operation,

said coding method being characterized in that during storage (C0) of the set of transform operations associated with said selected prediction mode, the number of transform operations contained in said set is different from the number of operations of transforms contained in a set of transform operations that is stored in association with at least one other predetermined prediction mode of said plurality.

The coding method according to claim 1, wherein for at least two sets of transform operations stored respectively in association with two prediction modes of said plurality of predetermined prediction modes, the number of transform operations in each of said two sets contains in common at least one identical transform operation.

An encoding method according to claim 1 or claim 2, wherein the set of transform operations associated with said selected prediction mode is stored in association with at least one other prediction mode of said plurality of predetermined prediction modes. .

An encoding method according to any one of claims 1 to 3, wherein the number of transform operations contained in a stored set in association with a prediction mode associated with a vertical or horizontal prediction direction is greater than the number of transform operations contained in a stored set in association with a prediction mode associated with an oblique prediction direction.

The encoding method according to any one of claims 1 to 3, wherein the number of transform operations contained in a stored set in association with a prediction mode, for which the prediction is calculated by averaging more than two pixels. an edge of the current block is greater than or equal to the number of transform operations contained in a stored set in association with any other prediction mode.

The coding method according to any one of claims 1 to 3, wherein the number of transform operations contained in a stored set in association with a prediction mode that has been previously selected, in said plurality of prediction modes. predetermined, as the most probable prediction mode, is greater than the number of transform operations contained in a stored set in association with a prediction mode that has not been previously selected as the most probable prediction mode .

7. Device (CO) encoding at least one image (IC _j ) cut into blocks, comprising a processing circuit (CT_C) which, for a current block (B,) to be encoded of said image, is arranged for:

applying a transform operation to the data of said residue block, said transform operation belonging to a set of operations of a transform that is previously stored in association with the selected prediction mode,

coding the data obtained as a result of said transform operation,

said coding device being characterized in that the processing circuit is arranged to store the set of transform operations associated with said selected prediction mode, the number of transform operations contained in said set being different from the number of operations of a transform contained in a set of transform operations that is stored in association with at least one other predetermined prediction mode of said plurality.

8. A method for decoding a data signal (F) representative of at least one image (IC _j ) cut into blocks, implementing, for a current block (B,) to be decoded:

a determination (D3) in said data signal:

A prediction mode (MP _S ) of the current block to be decoded, said prediction mode belonging to a plurality of predetermined prediction modes,

a prediction (D4) of the current block according to the determined prediction mode (MP _S ),

an application (D7) of a transform operation to data representative of said residual block, said transform operation belonging to a set of transform operations that is previously stored (DO) in association with said determined prediction mode (MP _S )

a reconstruction (D8) of the current block using a predictor block obtained at the end of the prediction and data obtained following said transform operation,

said decoding method being characterized in that during storage (D0) of the set of transform operations associated with said determined prediction mode, the number of transform operations contained in said set is different from the number of transform operations contained in a set of transform operations that is stored in association with at least one other predetermined prediction mode of said plurality.

A decoding method according to claim 8, wherein for at least two sets of transform operations respectively stored in association with two prediction modes of said plurality of predetermined prediction modes, the number of transform operations in each of said two sets contains in common at least one identical transform operation.

A decoding method according to claim 8 or claim 9, wherein the set of transform operations associated with said determined prediction mode is stored in association with at least one other prediction mode of said plurality of predetermined prediction modes. .

1 1. A decoding method according to any one of claims 8 to 10, wherein the number of transform operations contained in a stored set in association with a prediction mode associated with a vertical or horizontal prediction direction is greater than the number of transform operations contained in a stored set in association with a prediction mode associated with an oblique prediction direction.

The decoding method according to any one of claims 8 to 10, wherein the number of transform operations contained in a set associated with a prediction mode, for which the prediction is calculated by averaging more than two pixels. an edge of the current block is greater than or equal to the number of transform operations contained in a stored set in association with any other prediction mode.

The decoding method according to any one of claims 8 to 10, wherein the number of transform operations contained in a stored set in association with a prediction mode that has been previously selected, in said plurality of predetermined prediction modes, as the most probable prediction mode, is greater than the number of transform operations contained in a stored set in association with a prediction mode that has not been previously selected as the most likely prediction mode.

14. Device for decoding a data signal (F) representative of at least one image (IC _j ) cut into blocks, comprising a processing circuit (CT_D) which, for a current block (B,) to be decoded, is arranged for:

determining, in said data signal:

A prediction mode of the current block to be decoded, said prediction mode belonging to a plurality of predetermined prediction modes,

predict the current block according to the determined prediction mode,

applying a transform operation to data representative of said residue block, said transform operation belonging to a set of transform operations that is previously stored in association with said determined prediction mode (MP _S ),

reconstructing the current block using a predictor block obtained at the end of the prediction and data obtained following said transform operation,

said decoding device being characterized in that the processing circuit is arranged to store the set of transform operations associated with said determined prediction mode, the number of transform operations contained in said set being different from the number of operations of a transform contained in a set of transform operations that is stored in association with at least one other predetermined prediction mode of said plurality.

A computer program comprising program code instructions for performing the steps of the decoding method according to any one of claims 8 to 13, when said program is executed on a computer.