JPH0730899A - Hierarchical motion vector detection system - Google Patents

Abstract

PURPOSE:To improve the coding efficiency by measuring a difference of a motion vector of blocks of a proper number adjacent to an object unit block so as to detect a motion vector at a layer with high resolution. CONSTITUTION:A layer with lowest resolution is selected in a picture input signal of a current frame and an output of a multiplexer 103 for a decoded picture signal of a preceding frame and a motion vector detector 104 detects a motion vector for each unit block. An initial bias is given to the detected motion vector depending on magnification of the picture and a low-order layer is sequentially selected and the motion vector is detected for each unit block and the motion vector for each unit block of the current frame is detected finally. The value is set as an initial bias and the motion vector is detected at a layer with higher resolution and detected sequentially for lower layers and the motion vector is finally detected.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image communication device such as a TV telephone connected to a digital line and, more particularly, to moving image coding using predictive coding by motion compensation and transform coding by orthogonal transform. The present invention relates to a motion vector detection method in.

[0002]

2. Description of the Related Art Conventionally, as a technique of a motion vector detecting method using a hierarchy, "Motion amount detecting method in moving image using hierarchical pixel information", IEICE Transactions, J.
72-D-II, No. There are three examples. In the technique described in "Motion amount detection method in moving image using hierarchical pixel information", the frequency band is reduced to 1/2 from the image input signal.
Then, an image with a sampling frequency halved is generated, and by repeating the same processing, hierarchical images of lower resolution can be sequentially generated from the original image. The detection of the motion vector is sequentially performed from the upper layer having a lower resolution to the lower layer having a higher resolution. By keeping the number of pixels in the block to be detected constant in each layer, the detection range is adaptively switched from large to small and subdivided, thus improving the spatial resolution of the motion vector and expanding the detection range of the motion vector. Planned.

[0003]

In the conventional method, the motion vector is detected by dividing the entire frame of the motion vector detection into uniform unit blocks. However, in a normal image, a part with a large amount of motion and a part with a small amount of motion may coexist in the same frame, or there may be many objects with small motion in a unit block. When divided into unit blocks that are uniform over the entire area, the accuracy of the predicted image is low in motion vector detection in a unit block that has a large number of pixels, and there is motion in an area that spans multiple blocks. When different motion vectors are detected in a block, the discontinuity of the prediction image generated by motion compensation at the block boundary becomes significant, the amount of code generated by the prediction error increases, and the coding efficiency also deteriorates. As a result, there is a problem that block discontinuity remains in the reproduced image depending on the transmission rate. Further, when the number of pixels forming a unit block is reduced, the number of times of matching of blocks for detecting a motion vector unnecessarily increases, which makes it impossible to perform quick encoding in a moving image encoding apparatus that requires real-time processing. was there.

[0004]

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and a plurality of pixels of a predetermined number adjacent to a digitized image input signal are grouped together to form a coding unit block. , For each unit block, the positional relationship with the current frame is detected from the difference information with the corresponding block of the previous frame, that is, the motion vector is detected, and a prediction image is generated by compensation with the detected motion vector, and the current frame image and the prediction image In the motion vector detecting means of the moving picture coding apparatus for calculating the prediction error from the difference of the above, and having a sampling means for dividing the resolution of the image input signal into a plurality of layers It is a hierarchical motion vector detection method that sequentially detects motion vectors in a hierarchy, and an appropriate number of blocks adjacent to the target unit block. By measuring the difference between the click of a motion vector,
Provided is a hierarchical motion vector detection method in which a motion vector is sequentially detected in a layer having a high resolution for a target unit block when an absolute value of a motion vector difference is larger than a predetermined threshold value.

Further, according to the present invention, a predetermined number of pixels adjacent to each other of a digitized image input signal are collectively set as an encoding unit block, and each unit block is represented based on difference information from a corresponding block of a previous frame. Positional relationship with a frame, that is, a motion vector is detected, a prediction image is generated by compensation with the detected motion vector, a prediction error is calculated from the difference between the current frame image and the prediction image, and coding is performed. In the motion vector detecting means of the apparatus, a hierarchical motion vector detecting method is provided which has a sampling means for dividing the resolution of the image input signal into a plurality of layers and sequentially detects the motion vector from a layer having a low resolution to a layer having a high resolution. Therefore, the difference in brightness between adjacent pixels in an appropriate direction in the target unit block is measured, and the maximum difference in brightness between pixels is measured. Absolute value, if greater than a threshold which is predetermined to provide a hierarchical motion vector detecting method for detecting a motion vector in a sequential high resolution hierarchy for the target unit blocks.

[0006]

As a hierarchical image obtained by dividing an image input signal into a plurality of resolutions, a motion vector is sequentially detected from a layer having a low resolution to a layer having a high resolution, and an appropriate number of units adjacent to a unit block for detecting the motion vector are provided. A motion vector is adaptively detected in a layer with high resolution from the absolute value of the difference between block motion vectors. Further, the motion vector is adaptively detected in a layer having a high resolution from the absolute value of the difference in luminance between the pixels in the adjacent appropriate directions in the unit block for detecting the motion vector.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a first embodiment of the present invention, in which 100 is a digitized image input signal,
01 and 102 are a down sampler and an up sampler which divide the resolution of the image input signal into a plurality of layers, 103
Is a multiplexer for selecting a hierarchy of separated resolutions. Reference numeral 104 is a motion vector detector that detects a motion vector for each unit block, and 105 is a motion vector value input hierarchy determiner that determines a hierarchy for motion vector detection based on the value of the detected motion vector, and controls the 103 multiplexer. Reference numeral 106 denotes a subtracter that generates a prediction error image from a current frame image and a motion-compensated prediction image.
Reference numeral 10 is a control unit of the encoder. The encoder is a switch 1 for switching between intraframe coding and interframe coding.
07, the orthogonal transformer 108, and the quantizer 109. Reference numeral 115 denotes a local decoder portion for generating a predicted image, which includes an inverse quantizer 111, an orthogonal inverse transformer 112,
It is composed of a frame memory 114 having a motion compensation function and an adder 113 for obtaining a decoded image.

Next, the operation in such a configuration will be described. In FIG. 1, a digitized image input signal 100 is divided into a plurality of layers by a down sampler 101 and an up sampler 102 and input to a multiplexer 103. Among the methods of dividing the resolution into a plurality of layers, as a downsampling process, a method of performing sub-sampling after applying a low pass filter is generally used.
According to the method of calculating the average between pixels to obtain a low resolution image, the phase shift of half a pixel does not occur. Further, as an example of obtaining an enlarged image in the upsampling process, there is a method as shown in FIG.

First, a layer having the lowest resolution is selected from the outputs of the multiplexer 103 for the image input signal of the current frame and the decoded image signal of the previous frame, and the motion vector detector 104 detects the motion vector for each unit block. The detected motion vector is given an initial deviation according to the enlargement ratio of the image, the lower hierarchy is sequentially selected to detect the motion vector for each unit block, and finally the motion for each unit block of the current frame is performed. Vector values are detected. Motion vector detection will be described with reference to FIG. If one unit block is focused on, the motion vector detected in the lowest resolution layer captures a rough motion, and Vh is shown in the figure as a value considering the enlargement ratio of the image.
Shown as +2. Using this value as the initial deviation, the motion vector is detected in a layer with a higher resolution and Vh in the figure is detected.
It is shown as -1. Detection is performed sequentially in the lower hierarchy,
Eventually, the motion vector V will be detected.

In the local decoder unit 115, the error image is decoded by the inverse quantizer 111 and the inverse orthogonal transformer 112, and this is combined with the motion-compensated image of the previous frame in the frame memory 114 to obtain the predicted image of the current frame. Is generated. The subtractor 106 generates a prediction error image from the difference between the current frame and the predicted image, and the orthogonal transformer 108 and the quantizer 109 encode it and output it as a quantized output 116 of the transform coefficient together with the quantization characteristic 117.

In the first embodiment of the present invention, first, after detecting the motion vector in the layer with the lowest resolution, the difference between the motion vectors between adjacent unit blocks is calculated, and the absolute value of the difference is calculated from a preset value. When the value is large, that is, when the motion direction and the motion amount between the unit blocks are significantly different, the motion vector is detected only in the corresponding block in a hierarchy with a higher resolution. If the number of pixels included in this unit block for motion vector detection is fixed, the range of the detection target is divided into small regions, and the resolution of detection is increased. In the present embodiment, the process is further repeated to detect the motion vector up to the image of the layer in which the original image is enlarged, and the motion vector can be detected with the accuracy of a half pixel unit. A state in which blocks are divided to sequentially detect motion vectors will be described with reference to FIG. If the unit block for detecting the motion vector is MBN (i, j) in the figure, the motion vector Vh + 2 obtained in the low-resolution layer has a difference from the motion vector of the right adjacent unit block MBN (i + 1, j). Since it is large, detection is performed in a layer with a higher resolution. However, since the motion vector of the d block among the blocks of a, b, c, and d that make up the MBN (i, j) has a large difference from the motion vector of the surrounding unit blocks, it is possible to further increase the resolution in the hierarchy. Motion vector detection is performed.
Since the method of detecting the motion vector by lowering the hierarchy adaptively only when the direction of motion and the amount of motion of the block to be detected are significantly different from those of the surrounding blocks, there is a large amount of motion in the block. Even when a part and a part with a small amount of movement are mixed, or when there are many objects with small movement in a unit block,
The motion vector can be properly expressed with the minimum block division.

Another embodiment of the present invention will be described with reference to FIG. A motion vector is detected from the digitized image input signal and the prediction image of the previous frame, a prediction image by motion compensation is generated, and a prediction error from the input image of the current frame is encoded and output. The image luminance value input layer determiner 205 adaptively selects a layer of an image for which a motion vector is detected. When a unit block for detecting a motion vector includes a boundary between regions or the like, there is a high possibility that the motion will occur independently or that it is a subject, and it is necessary to detect the motion in detail. Since the differential value of the luminance can be used for extracting such a region, the determination of the hierarchy for detecting the motion vector is performed by obtaining the difference between the luminance values of the adjacent pixels in the unit block of the input image of the current frame, When it is determined from the absolute value and is larger than a preset threshold value, motion vector detection in a higher resolution hierarchy is performed for the block. This makes it possible to detect the movement of the object in the vicinity of the area in detail.

In each of the above-mentioned embodiments, the hierarchical structure has two layers for low resolution and one layer for high resolution, but it can be applied to any hierarchical structure.

[0014]

As described above, according to the configuration and method of the present invention, a portion having a large amount of movement and a portion having a small amount of movement are mixed in the same frame, or a small movement occurs in a unit block. Even if there are many objects or there is motion in a region that spans multiple blocks, the motion vector can be properly expressed with a very small number of block divisions, so the accuracy of the prediction image generated by motion compensation is high. It is high, the code amount generated due to the prediction error is very small, and the coding efficiency is improved. As a result,
It becomes possible to control the amount of generated code, block discontinuity does not remain in the reproduced image even when the transmission rate is low, and a very natural and excellent image quality can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of the present invention.

FIG. 2 is a block diagram showing a second embodiment of the present invention.

FIG. 3 is an explanatory diagram showing an example of downsampling.

FIG. 4 is an explanatory diagram showing an example of upsampling.

FIG. 5 is an explanatory diagram of a hierarchical detection method of motion vectors.

FIG. 6 is an explanatory diagram of hierarchical motion vector detection in this embodiment.

[Explanation of symbols]

 100 Image Input Signal 101 Downsampler 102 Upsampler 103 Multiplexer 104 Motion Vector Detector 105 Motion Vector Value Input Hierarchical Determinator 106 Subtractor 107 Changeover Switch 108 Orthogonal Transformer 109 Quantizer 110 Encoding Control Unit 111 Inverse Quantizer 112 Orthogonal inverse transformer 113 Adder 114 Frame memory with motion compensation function 115 Local decoder unit 116 Quantization output of transform coefficient 117 Quantization characteristic output 118 Intra / inter determination 119 Motion vector value 205 Image luminance value input hierarchical determination device

Claims (2)

[Claims] 1. A plurality of pixels of a predetermined number which are close to each other in a digitized image input signal are collectively set as an encoding unit block, and for each unit block, a difference from the corresponding block of the preceding frame to the current frame is determined. A motion of a moving picture coding apparatus that detects a positional relationship, that is, a motion vector, generates a predicted image by compensation with the detected motion vector, calculates a prediction error from a difference between the current frame image and the predicted image, and performs coding. In the vector detection means, there is a sampling means for dividing the resolution of the image input signal into a plurality of layers, which is a hierarchical motion vector detection method for sequentially detecting a motion vector from a low resolution layer to a high resolution layer, The difference between the motion vectors of an appropriate number of blocks adjacent to the target unit block is measured, and the absolute value of the difference between the motion vectors is estimated. Is greater than a defined threshold, hierarchical motion vector detecting method, characterized by performing the detection of the motion vector in a sequential high resolution hierarchy for the target unit blocks. 2. A plurality of pixels of a predetermined number adjacent to each other in a digitized image input signal are collectively set as an encoding unit block, and for each unit block, difference information from a corresponding block of a previous frame is compared with a current frame. A motion of a moving picture coding apparatus that detects a positional relationship, that is, a motion vector, generates a predicted image by compensation with the detected motion vector, calculates a prediction error from a difference between the current frame image and the predicted image, and performs coding. In the vector detection means, there is a sampling means for dividing the resolution of the image input signal into a plurality of layers, which is a hierarchical motion vector detection method for sequentially detecting a motion vector from a low resolution layer to a high resolution layer, The difference in brightness between pixels in adjacent appropriate directions in the target unit block is measured, and the maximum absolute value of the difference in brightness between pixels is calculated in advance. Is larger than order threshold, hierarchical motion vector detecting method, characterized by performing the detection of the motion vector in a sequential high resolution hierarchy for the target unit blocks.