JP2012186760A - Video decoding device, video decoding method, and video decoding program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the program that, when a motion vector of a block at the same position as that in another picture is used as a candidate of a predictive motion vector, since it is necessary to record the motion vector of another picture that has already been decoded in a memory of a decoding device, the data amount of motion vectors becomes huge, and an additional memory becomes indispensable, so that cost increases.SOLUTION: A motion vector compression/decompression unit 209, before starting decoding of the next picture, extracts motion vectors recorded in a memory, and selects one motion vector from among plural motion vectors in a predetermined range as a representative value of the predetermined range, thereby reducing the storage capacity for motion vectors occupied in the memory. The predetermined range is designated at the time of encoding, and data structure stored as information is also defined in an encoded bit stream, so that reduction of motion vectors similar to that in encoding is also enabled in decoding.

Description

  The present invention relates to a moving picture decoding technique, and more particularly to a moving picture decoding technique using motion compensated prediction.

  In a moving picture coding method that divides a picture into rectangular blocks, represented by MPEG (Moving Picture coding Experts Group), and performs motion estimation and compensation between blocks, the coding of the motion vector generated in each block In order to reduce the amount, a prediction process is performed on the motion vector. In MPEG-2, the motion vector detected in units of macroblocks is taken as the difference from the motion vector of the macroblock encoded immediately before, and the amount of code is reduced by encoding the difference vector. Yes. In MPEG-4 AVC / H.264, by using the fact that motion vectors have a strong correlation with the motion vectors of neighboring neighboring blocks, prediction from neighboring neighboring blocks is performed, and the difference vector is encoded. The code amount is reduced. Specifically, as shown in FIG. 1, a median value is calculated from motion vectors of neighboring neighboring blocks, and a motion vector prediction is realized by taking a difference from the median value. However, when the shape of the encoding target block and the adjacent block are different as shown in FIG. 2, when there are a plurality of adjacent blocks on the left side, the top block among them is shown, and the plurality of adjacent blocks are shown above. In some cases, the leftmost block is a prediction block, and if the block to be encoded is divided by 2N × N pixels or N × 2N pixels as shown in FIG. A prediction block is determined according to the size as shown in FIG. 3, and prediction is performed from the motion vector.

ISO / IEC 13818-2 Information technology-Generic coding of moving pictures and associated audio information: Video ISO / IEC 14496-10 Information technology-Coding of audio-visual objects-Part 10: Advanced Video Coding

  In the conventional motion vector prediction method, since there is only one motion vector for prediction, there is a problem that if the prediction is not successful, the difference between the motion vectors becomes large and the generated code amount increases. In order to solve this problem, a new motion vector prediction method has been studied in the standard work of moving picture coding in ISO / IEC and ITU-T. This method is a generated code when each of the motion vectors of neighboring blocks that have been encoded and the motion vector of a block at the same position of another picture that has been encoded are applied as prediction motion vector candidates. It is a standard for evaluating the quantity. FIG. 4 shows an example of adjacent blocks that are candidates for the motion vector predictor. FIG. 4 (a) is an example of an adjacent block in the same picture. One of the blocks A0 to A3 adjacent to the left side of the block to be processed is selected as the left adjacent block, and the block B0 adjacent to the upper side is selected. One of B5 is selected as the upper adjacent block. FIG. 4B is an example of a block at the same position in another picture with different time. The motion vectors of these three blocks are used as prediction motion vector candidates, the motion vector with the smallest generated code amount is selected as the prediction motion vector, and additional information related to the adjacent block for which the motion vector is selected is encoded and transmitted if necessary. To do. However, since it is necessary to record the motion information of another already encoded picture in the memory of the encoding device, it is indispensable to add the memory, and there is a problem that the cost increases. Although motion information is calculated by motion detection for each prediction block, even if the size of the prediction block for motion detection is adaptively changed, it is necessary to assume motion detection in the prediction block of the minimum size. Needs to store motion information in minimum size units. Therefore, even when the prediction block is large, one motion information of the prediction block is not stored in the memory, but the same motion information is stored in all the minimum sizes constituting the prediction block. On the other hand, in order to obtain prediction motion vector candidates from a block at the same position in another picture, it is easier to access the memory if it is recorded in the memory in the minimum size unit of the prediction block even if motion information is duplicated. There are also advantages. Therefore, if motion information is stored in the memory according to the size of the prediction block, the use capacity of the motion information for the memory is reduced, but the conversion from the prediction block address to the memory address becomes complicated and the access efficiency decreases. The problem which ends.

  In view of the above points, the present invention relates to motion information of another picture that has already been decoded, in particular, a motion vector, and selects one motion vector from a plurality of motion vectors within a predetermined range as a representative value of the predetermined range. It is an object of the present invention to provide a technique for reducing the storage capacity of motion vectors in the memory. In the present invention, a prediction block having a large area occupying within a predetermined range is determined, and the motion vector is used as a representative value, thereby improving the accuracy of a prediction motion vector obtained from another picture and encoding efficiency. Improve.

In order to solve the above problems, the present invention provides the following apparatus, method, and program.
(1) A moving picture decoding apparatus for decoding a bitstream in which the moving picture is encoded using a motion vector in units of blocks obtained by dividing each picture of the moving picture,
Using motion information including a motion vector decoded in units of blocks within a predetermined range from the bitstream, and using a parameter indicating the size of a prediction block to which the block within the predetermined range decoded from the bitstream belongs Determining a prediction block having a large area occupying the predetermined range, storing the determined motion vector of the prediction block in a memory as a representative value of the predetermined range, and at the time of decoding the next picture, A motion vector compression / decompression unit that selects a representative value of the motion vector from the memory according to the position of the prediction block;
A predicted motion vector selection unit that decodes a predicted motion vector using a representative value of the motion vector selected by the motion vector compression / decompression unit as a candidate for a predicted motion vector;
A moving picture decoding apparatus comprising:
(2) In the video decoding device according to (1) above,
The prediction block is a rectangular pixel area having one or a plurality of the divided blocks, and is divided so as to have the same motion information and a parameter representing the size of the prediction block. Decryption device.
(3) In the video decoding device according to (1) above,
In the divided block, a reference picture number and a prediction mode are stored as the motion information in addition to a motion vector, and the parameter indicating the size of the prediction block to which the divided block belongs is set as the size of the prediction block. A moving picture decoding apparatus, wherein a division mode indicating the number of divisions up to and a shape of the prediction block is stored.
(4) In the video decoding device according to (1) above,
The predetermined range is defined by a square rectangular array that equally divides the motion vector data array arranged two-dimensionally so as to correspond to the picture in the memory, and the number of blocks divided as the side length of the square rectangular array A video decoding device characterized in that it obtains a parameter representing the above from a bit stream.
(5) In the video decoding device according to (1) above,
The motion vector compression / decompression unit selects a motion vector of the prediction block as a representative value by prioritizing the previous prediction block in decoding order when a plurality of prediction blocks remain as selection candidates in the predetermined range. A moving picture decoding apparatus characterized by:
(6) In the video decoding device according to (1) above,
The motion vector compression / decompression unit represents a motion vector of the prediction block by giving priority to a prediction block located near the center of the predetermined range when a plurality of prediction blocks remain as selection candidates in the predetermined range. A moving picture decoding apparatus characterized by selecting as a value.
(7) In the video decoding device according to (1) above,
The motion vector compression / decompression unit has the largest predicted block size to which the plurality of blocks belong based on motion information of a plurality of blocks arranged in the center of the predetermined range and a parameter representing the size of the predicted block. A moving picture decoding apparatus, wherein a motion vector of a block is selected as a representative value.
(8) In the video decoding device according to (1) above,
The motion vector compression / decompression unit previously determines a minimum determination range for performing determination of motion vector selection based on the size of the predetermined range, and a prediction block in the predetermined range is larger than the minimum determination range. In some cases, the motion decoding apparatus is characterized in that the motion vector of the prediction block is selected as a representative value, and the determination process is terminated.
(9) A moving picture decoding method for decoding a bitstream in which the moving picture is encoded using a motion vector in units of blocks obtained by dividing each picture of the moving picture,
Using motion information including a motion vector decoded in units of blocks within a predetermined range from the bitstream, and using a parameter indicating the size of a prediction block to which the block within the predetermined range decoded from the bitstream belongs Determining a prediction block having a large area occupying the predetermined range, storing the determined motion vector of the prediction block in a memory as a representative value of the predetermined range, and at the time of decoding the next picture, And a motion vector compression / decompression step for selecting a representative value of the motion vector from the memory according to the position of the prediction block;
A predicted motion vector selection step of decoding a predicted motion vector using a representative value of the motion vector selected in the motion vector compression / decompression step as a predicted motion vector candidate;
A moving picture decoding method comprising:
(10) A moving picture decoding program for causing a computer to execute an operation of decoding a bitstream in which the moving picture is encoded using a motion vector in units of blocks obtained by dividing each picture of the moving picture,
Using motion information including a motion vector decoded in units of blocks within a predetermined range from the bitstream, and using a parameter indicating the size of a prediction block to which the block within the predetermined range decoded from the bitstream belongs Determining a prediction block having a large area occupying the predetermined range, storing the determined motion vector of the prediction block in a memory as a representative value of the predetermined range, and at the time of decoding the next picture, And a motion vector compression / decompression step for selecting a representative value of the motion vector from the memory according to the position of the prediction block;
A predicted motion vector selection step of decoding a predicted motion vector using a representative value of the motion vector selected in the motion vector compression / decompression step as a predicted motion vector candidate;
Is a moving picture decoding program for causing a computer to execute.

  According to the present invention, with respect to a motion vector of another picture that has already been decoded, one motion vector is selected as a representative value of the predetermined range from a plurality of motion vectors within the predetermined range, and stored in the memory. The storage capacity of motion vectors in the memory can be reduced.

It is a figure explaining how to obtain a predicted motion vector from conventional neighboring blocks. It is a figure explaining the calculation method of the prediction motion vector in case the shape of the conventional adjacent block differs. It is a figure explaining how to obtain a prediction motion vector when the shape of a conventional encoding target block is 2N × N or N × 2N. In motion vector prediction, it is a figure for demonstrating the operation | movement which selects the candidate of a motion vector predictor from the adjacent block of a periphery, and the block of the same position of another picture. It is a block diagram which shows the structure of the moving image encoder which comprised the selection method of the motion vector in an Example. It is a block diagram which shows the structure of the moving image decoding apparatus which comprised the selection method of the motion vector in an Example. It is a figure for demonstrating the definition of the encoding block in an Example. It is a figure for demonstrating the kind of shape of the prediction block in an Example. It is a figure for demonstrating the kind of shape of the CU partition in an Example. It is a figure explaining the preservation | save format of the motion vector selected by the selection method of the motion vector in an Example. It is a figure which shows the syntax pattern of the bit stream which determines whether the motion vector selection method in an Example is performed. It is a block diagram which shows the detailed structure for demonstrating operation | movement of the motion vector compression / decompression part of the moving image encoder which provided the motion vector selection method in an Example, and a moving image decoder. It is a figure for demonstrating the position of the prediction block used as the process target of the selection method of the motion vector in an Example. It is a block diagram which shows the detailed structure for demonstrating operation | movement of the motion vector compression part of the moving image encoder which provided the motion vector selection method in an Example, and a moving image decoder. It is a figure explaining an example of depth and partMode recorded on a memory in the minimum prediction block unit in the predetermined range set with the selection method of the motion vector in an example. 6 is a flowchart for explaining the operation of the first embodiment of the motion vector selection method; It is a figure explaining arrangement | positioning of the predetermined range set with the selection method of the motion vector in an Example, and the minimum prediction block inside it. It is a figure for demonstrating the structure of the block by which the largest encoding block defined with the selection method of the motion vector in an Example and the inside are divided | segmented. 6 is a flowchart for explaining an operation of two-division selection determination in the first embodiment of the motion vector selection method. It is a figure for demonstrating an example of the division | segmentation inside the minimum coding block defined with the selection method of the motion vector in an Example. FIG. 6 is a diagram for explaining the position of a minimum prediction block selected within a predetermined range in the first embodiment of the motion vector selection method. 4 is a flowchart for explaining an operation of 4-division selection determination in the first embodiment of the motion vector selection method; It is a table which defines the index of the division | segmentation mode used with the selection method of the motion vector in an Example. FIG. 10 is a diagram for explaining the position of the minimum prediction block selected within a predetermined range in the second embodiment of the motion vector selection method. FIG. 10 is a table for designating a position of a minimum coding block within a predetermined range and a position of a minimum prediction block selected from a division mode of the minimum coding block in the second embodiment of the motion vector selection method. FIG. 10 is a diagram for explaining the position of a minimum prediction block to be determined within a predetermined range in the third embodiment of the motion vector selection method. 12 is a flowchart for explaining an operation of selecting a representative value from four minimum prediction blocks within a predetermined range of the third embodiment of the motion vector selection method. FIG. 20 is a diagram for describing an example of selecting a minimum prediction block when the first embodiment is applied from a predetermined divided range in the fourth embodiment of the motion vector selection method. 10 is a flowchart for explaining the operation of a fourth embodiment of the motion vector selection method.

  A video encoding device and a video decoding device that are embodiments of the present invention will be described. FIG. 5 is a block diagram illustrating a configuration of the moving image encoding apparatus according to the embodiment. The motion vector detecting unit 101 that detects a motion vector by performing matching between pictures in units of blocks, and subtracts an image to be encoded and a predicted image. A subtractor 102 that generates a residual signal, an orthogonal transform and quantization unit 103 that performs orthogonal transform and quantization on the residual signal, and a variable-length encoding unit 104 that performs entropy encoding, a transform Inverse quantization / inverse orthogonal transform unit 105 that returns the original residual signal by inverse quantization and inverse orthogonal transform of the signal, motion compensation unit 106 that generates a prediction image using a motion vector, and adaptively multiplying the weight coefficient A weighted prediction unit 107 that generates a predicted image, an adder unit 108 that generates a decoded image by adding the predicted image and the residual signal, a deblocking filter unit 109 that reduces block distortion due to encoding, Coding A memory 110 for storing information on the recorded image, a motion vector selector 111 that selects an optimal motion vector predictor based on motion information of an already encoded image, and a motion vector recorded in the memory 110 is selected. A motion vector compression / decompression unit 112 is provided.

  FIG. 6 is a block diagram showing a configuration of a moving picture decoding apparatus according to an embodiment of the present invention corresponding to the moving picture encoding apparatus of FIG. 5, and a variable length decoding unit 201 that decodes a bit stream and outputs various information. An inverse quantization / inverse orthogonal transform unit 202, a motion compensation unit 203, a weighted prediction unit 204, an addition unit 205, a deblocking filter unit 206, a memory 207, a predicted motion vector selection unit 208, and a motion vector compression / decompression unit 209. It is to be prepared. Each unit 202 to 209 of the video decoding device corresponds to a decoding process provided inside the video encoding unit, and therefore corresponds to each unit 105 to 112 of the video encoding device. It has a function.

  The motion vector selection method for reducing the storage capacity of the motion vector in the memory according to the embodiment is the motion vector compression / expansion unit 112 of the video encoding device and the motion vector compression / expansion unit 209 of the video decoding device. To be implemented. The motion vector compressing / decompressing unit 112 of the moving image encoding apparatus finishes encoding one picture, records a decoded picture used as a reference picture, and simultaneously encodes a motion vector provided for each block of the picture. Each predetermined range designated at the time of conversion is extracted from the memory 110, representative values are selected from a plurality of motion vectors included in the predetermined range, and recorded in the memory 110 again. When referring to a motion vector of a reference picture as a motion vector predictor candidate in the motion vector predictor selection unit 111, an address of a predetermined range corresponding to a block at the same position as the block to be processed is calculated, and the corresponding predetermined range from the memory 110 Are read out. The motion vectors of adjacent blocks around the same picture that have already been encoded are input to the predicted motion vector selection unit 111, and the motion vector with the smallest generated code amount is selected as the predicted motion vector. On the other hand, the motion vector compression / decompression unit 209 of the moving picture decoding apparatus similarly records a picture decoded from the bit stream as a reference picture for decoding the next picture, and at the same time prepares for a block unit of the picture. A motion vector is extracted from the memory 207 for each predetermined range, a representative value is selected from a plurality of motion vectors included in the predetermined range, and is recorded in the memory 207 again. Here, the predetermined range is calculated from information indicating the predetermined range decoded from the bitstream. When the motion vector of the reference picture is referred to as a motion vector predictor candidate by the motion vector predictor selection unit 208, an address of a predetermined range corresponding to a block at the same position as the block to be processed is calculated, and the corresponding predetermined range from the memory 207 is calculated. Are read out. The motion vectors of adjacent blocks around the decoded same picture are input to the predicted motion vector selection unit 111, and the motion vector with the smallest generated code amount is selected as the predicted motion vector.

  In this way, the predetermined range is specified at the time of encoding, and the data structure stored as information is also defined in the encoded bit stream, so that the motion vectors can be reduced in the same way as in encoding. It becomes. Since the predetermined range is larger than the minimum size of the prediction block in which motion information is stored and there is no motion vector that originally existed, the encoding efficiency may be lower than the original motion vector when selected as a prediction motion vector. In the present invention, it is possible to improve the accuracy of a prediction motion vector obtained from another picture by determining a prediction block having a large area occupying within a predetermined range and using the motion vector as a representative value. , Avoiding a decrease in encoding efficiency.

In the following embodiment, details of a motion vector selection method that is commonly performed in the motion vector compression / expansion unit 112 of the video encoding device and the motion vector compression / expansion unit 209 of the video decoding device will be described.
[Example]
Before describing an embodiment of a motion vector selection method according to the present invention, terms used in this embodiment will be defined.
Definition of coding block In this embodiment, as shown in FIG. 7, the screen is equally divided into square rectangular blocks of the same size. This block is called a coding block and is the basis of processing when performing coding and decoding. In order to optimize the encoding process according to the texture in the screen, the encoded block can be divided into four to make the encoded block small in block size. An encoded block in which the screen shown in FIG. 7 is divided into equal sizes is referred to as a maximum encoded block, and the inside of which is divided into four according to the encoding conditions is collectively referred to as an encoded block. An encoded block having a minimum size in which the encoded block cannot be further divided into four is referred to as a minimum encoded block.
・ Definition of prediction block When performing motion compensation by dividing the screen into blocks, the smaller the motion compensation block size, the more detailed prediction can be performed. It adopts a mechanism to select the optimal one and divide the inside of the block to perform motion compensation. A block that performs this motion compensation is called a prediction block. The prediction block is used only when the inside of the minimum coding block is divided, and according to motion compensation, the case where the inside of the minimum coding block is regarded as one block without being divided is maximized and divided into two in the horizontal or vertical direction. Can be divided into four parts by horizontal and vertical equal divisions. A mode corresponding to the division type is defined according to the size after the division, and is shown in FIG. The numbers inside the rectangles in FIG. 8 represent the numbers of the divided prediction blocks. In order to manage the prediction block in the minimum coding block, numbers starting from 0 are assigned to the prediction block in the zigzag scan order.
・ Definition of CU partition Even if the block size of motion compensation is made smaller for a coding block of a size larger than the minimum coding block, more detailed prediction can be performed. A partition for motion compensation having a function equivalent to that of the prediction block applied to is defined. This division is called a CU (Coding Unit: synonymous with coding block) partition. Fig. 9 shows an example of partitioning by CU partition. Similar to the prediction block, motion detection / compensation is performed on the block obtained by dividing the encoded block into two in the horizontal or vertical direction. When the prediction block is divided into four parts by horizontal and vertical equal divisions, it is synonymous with the case where the coded block is divided into four parts and motion compensation is performed for each divided coded block. Not. The numbers inside the rectangles in FIG. 9 represent the numbers of the blocks divided by the CU partition.

  As described above, when motion detection / compensation is performed on an encoded block, the unit of motion detection / compensation is the size of the encoded block itself, the size obtained by dividing the encoded block into two by the CU partition, and 4 encoded blocks. Each divided small block, and further dividing the small block as an encoded block in the same manner, becomes a block of a size that can be obtained recursively up to the minimum size of the prediction block. Hereinafter, unless otherwise specified, the unit for performing motion detection / compensation is a “prediction block” regardless of the size.

  Next, recording of motion information such as motion vectors detected by motion detection / compensation will be described. Although motion detection / compensation is performed in units of prediction blocks, motion information such as detected motion vectors is not recorded in units of prediction blocks, but is recorded in units of minimum prediction blocks described above. That is, the same motion information is recorded in a plurality of minimum prediction blocks constituting the prediction block. Since the size of the prediction block varies depending on the encoding condition, additional information such as the position of the prediction block on the picture and the shape of the prediction block is required in addition to the detected motion vector in order to record in the memory in units of the prediction block. This complicates access for obtaining motion information. Therefore, although there is redundancy due to redundant recording of information, it is easier to access by dividing and recording into uniform minimum prediction block units. FIG. 10 shows an example in which a picture is divided into minimum prediction blocks, and the storage format of motion information recorded in the memories 110 and 207 will be described with reference to this figure. In FIG. 10, w represents the width of the picture, and h represents the height of the picture. If the width and height of the minimum prediction block are p, a picture is composed of w / p minimum prediction blocks in the horizontal direction and h / p minimum prediction blocks in the vertical direction. The motion information corresponding to is recorded in the memories 110 and 207 in the raster scan order represented by the thick solid line in FIG. Here, if a rectangular area surrounded by a thick line in FIG. 10 is a prediction block (showing a gray area in FIG. 10), motion information such as a motion vector is detected by motion detection in the prediction block. The detected motion information is recorded in the memories 110 and 207 corresponding to the minimum prediction block constituting the prediction block. When the start address of the motion information recording area in the memories 110 and 207 is the origin in the upper left in FIG. 10, the start position of the prediction block is indicated by a black circle in FIG. If there are y positions, the start address of the prediction block is represented by w / p × y + x. The motion information of the prediction block is recorded in the memories 110 and 207 corresponding to the minimum prediction block in the prediction block shown in gray from the minimum prediction block at the head position. In this way, the same motion information is recorded in the memories 110 and 207 in the smallest prediction block of the prediction blocks.

After the encoding of one picture is completed, the motion vector selection method according to the present invention is performed. The motion vector of the motion information recorded in the memories 110 and 207 is compressed, and the memory usage is reduced. The motion vector that has been subjected to the compression process is used as a motion vector candidate for a motion vector of a block at the same position of a reference picture with a different time in the prediction of the motion vector for encoding the next picture. The motion vector selection method according to the present invention extracts motion vectors of motion information recorded in the memories 110 and 207 for each predetermined range, selects a representative value from them, and records it again in the memories 110 and 207. . The predetermined range for performing the motion vector selection determination of the present invention is defined by the syntax described below.
(Syntax definition)
First, the syntax of a bit stream of a moving image that is encoded by a moving image encoding device including the motion vector selection method according to the present embodiment will be described.

  FIG. 11 shows a syntax pattern described in a sequence parameter set (hereinafter abbreviated as SPS) of a moving image bit stream. A first flag motion vector buffer comp flag indicating whether to reduce the memory usage by applying the motion vector selection method of the present invention over the entire bitstream is provided. When the motion vector buffer comp flag is false (0), the motion vector selection method of the present invention is not applied, and the motion information is recorded in the memory as it is. When the motion vector buffer comp flag is true (1), a second flag motion vector buffer comp ratio indicating a parameter representing the size of a predetermined range for actually performing the motion vector selection determination is set. The minimum size of motion compensation is a prediction block obtained by dividing the minimum coding block into four by horizontal and vertical equal divisions, and motion information is recorded in the memory for each prediction block. This prediction block is called a minimum prediction block. The size of the predetermined range for performing the motion vector selection determination according to the present invention is such that the minimum prediction blocks for the number represented by the motion vector buffer comp ratio are arranged in the vertical and horizontal directions, and when the number is MVBufferCompRatio, It is expressed by the following formula.

MVBufferCompRatio = 1 << motion vector buffer comp ratio
In other words, one minimum prediction block is selected from the minimum prediction block MVBufferCompRatio × MVBufferCompRatio in the predetermined range, and the motion vector of that block is recorded in the memory as a representative value of the predetermined range. Is reduced. The motion vector selection determination is performed in the raster scan order from the upper left to the lower right of the picture in units of the predetermined range. Here, there is a restriction on the motion vector buffer comp ratio setting so that the number of pixels by multiplying the number of vertical or horizontal pixels of the minimum prediction block by MVBufferCompRatio does not exceed the number of vertical or horizontal pixels of the maximum coding block. Is done. This is because the maximum coding block is a coding unit that equally divides the inside of a picture, and a predetermined range for performing motion vector selection determination cannot be set across this boundary. If the motion vector buffer comp ratio does not exist, it is set to “0”.

Next, the memory 110 and the motion vector compression / decompression unit 112 of the moving picture coding apparatus including the motion vector selection method according to the present invention, and the memory 207 and the motion vector compression / decompression unit 209 of the moving picture decoding apparatus The operation will be described.
(Selection of motion vector in encoding)
The operation of the motion vector selection method according to the present invention in a video encoding device that encodes a video bitstream based on the above syntax will be described.

  FIG. 12 is a diagram illustrating an internal configuration of the motion vector compression / decompression unit 112, and corresponds to a portion indicated by a thick frame line. The motion vector compression / decompression unit 112 has both functions of motion vector compression and decompression, and switching between compression and decompression is enabled by internal switches 305 and 306.

  First, the operation of motion vector compression processing will be described. When encoding of a picture to be encoded is completed, selection (compression) of a motion vector recorded in the memory 110 is started. First, the switches 305 and 306 of the motion vector compression / decompression unit 112 are set to the compression side. The switches 305 and 306 operate in conjunction with each other to switch between compression and expansion, and the switch 305 connects between the memory 110 and the motion vector compression unit 303. On the other hand, the switch 306 prohibits input of an address to the predetermined range position setting unit 302. Next, a parameter “motion vector buffer comp ratio” representing the size of a predetermined range in which the motion vector selection determination set in the compression encoding of the motion vector is performed is input to the predetermined range setting unit 301. The predetermined range setting unit 301 calculates the number MVBufferCompRatio of the predetermined range in the vertical and horizontal directions based on the above formula. This process may be set once at the start of encoding, recorded in the predetermined range setting unit 301, and not calculated every time a picture is encoded. In addition, when it is desired to change the number of the predetermined range, control may be performed so that an SPS header whose value is changed is inserted into the bit stream each time. Until the SPS header is updated, the number of the predetermined range set before that is recorded in the predetermined range setting unit 301 and may not be calculated every time the picture is encoded. In the position setting unit 302, a position corresponding to a predetermined range is calculated from the top address of the memory 110 in which the motion vector is recorded, and the address of the motion vector included in the predetermined range is designated. In the memory 110, motion vectors in a predetermined range designated by the predetermined range position setting unit 302 are extracted and output to the motion vector compression unit 303. In the motion vector compressing unit 303, one representative value is selected from the extracted motion vectors in a predetermined range and recorded again in the memory 110.

  Next, the operation of the motion vector expansion process will be described. Since the decompression process needs to refer to the motion vector of the prediction block of the reference picture existing at the same position as the prediction motion vector candidate for each prediction block, the motion vector compression is performed immediately before the encoding of the next picture is started. The switches 305 and 306 of the expansion unit 112 are set to the expansion side. The switch 305 connects between the memory 110 and the motion vector decompression unit 304. On the other hand, the switch 306 permits input of an address to the predetermined range position setting unit 302. The motion vector buffer comp ratio set at the time of encoding is input to the predetermined range setting unit 301. The predetermined range setting unit 301 calculates the number MVBufferCompRatio of the predetermined range in the vertical and horizontal directions based on the above formula. As an encoding control, when an SPS header is inserted into a bitstream to be encoded and the motion vector buffer comp ratio is changed, the MVBufferCompRatio is recalculated. Until then, the previously calculated MVBufferCompRatio is recorded in the predetermined range setting unit 301 and used, so this process need not be performed for each picture. The predetermined range position setting unit 302 specifies the address of the motion vector compressed and recorded in the memory 110 from the predicted block address of the picture to be encoded and the number of predetermined ranges calculated by the predetermined range setting unit 301. As shown in FIG. 13, the address of a prediction block (gray area in FIG. 13) defined for motion compensation within a coding block in a picture to be coded is represented by a black circle position. Assuming that the upper left of the picture is the head address, the head address of the prediction block is converted to an address of the minimum prediction block unit by dividing the pixel position of the black circle by the number of vertical or horizontal pixels of the minimum prediction block. The address of the predicted block is input to the predetermined range position setting unit 302. The number of vertical and horizontal MVBufferCompRatio in the predetermined range is the number that is thinned out to select the representative value of the motion vector. Address. The motion vector at the corresponding address is extracted from the memory 110 and sent to the motion vector decompression unit 304. The motion vector decompression unit 304 sends the extracted motion vector to the memory 110 as a predicted motion vector candidate from a prediction block existing at the same position of the reference picture.

  As described above, when the motion vector selection method according to the present invention is used, the first flag motion vector buffer comp flag and the second flag motion vector buffer comp ratio described in the SPS header of the bitstream are set. And encoded. If necessary, the SPS header can be updated and inserted into the bitstream.

When the motion vector selection method according to the present invention is configured as shown in FIG. 12 as an explanation of the operation, when motion vector detection / compensation is performed in units of prediction blocks and then motion vectors are selected, There is a possibility that the accuracy of the predicted motion vector is lowered. This is because the motion vector detected by the original motion detection is greatly affected by the motion vector compression processing. Even if you try to refer to the motion vector of the block adjacent to the left or above in the same picture, the motion vector compression causes the motion vector of the smallest prediction block that is one MVBufferCompRatio−1 apart to be the motion vector predictor candidate. As a result, the motion vector prediction does not operate sufficiently, and the generated code amount increases. In order to solve this problem, as shown in FIG. 14, a temporary storage memory 307 for temporarily recording a motion vector selected within a predetermined range is added inside the motion vector compression / decompression unit 112. . Since FIG. 14 only shows the motion vector compression processing, the switches 305 and 306 are omitted. The difference from FIG. 12 is that, in order to perform motion vector compression for each prediction block, the address of the prediction block is input, and the representative value of the selected motion vector is temporarily recorded. The memory memory 307 is added. Since it can be seen from the motion detection that the motion vector is recorded in the memory 110 within the predetermined range from the address of the prediction block, the predetermined range position is designated by the predetermined range position setting unit 302 each time, and the memory 110 A range of motion vectors is extracted and output to the motion vector compression unit 303. In the motion vector compression unit 303, one representative value is selected from the extracted motion vectors in a predetermined range. On the other hand, the predetermined range position setting unit 302 designates the recording position of the motion vector corresponding to the position of the predetermined range in the temporary storage memory 307, and records the representative value there. When the encoding of the picture to be encoded is completed, the representative value of the motion vector recorded in the temporary storage memory 307 is recorded in the memory 110. As described above, the motion vector compression / decompression unit 112 shown in FIG. 14 to which the temporary storage memory 307 is added can compress motion vectors in units of prediction blocks. It should be noted that the motion vector expansion processing is the same as the conventional one, and is therefore omitted.
(Prediction of motion vector in decoding)
The operation of the motion vector selection method according to the present invention in the video decoding device that decodes the encoded video bitstream based on the above-described syntax will be described. The basic concept is the same as that of the above-described video encoding device, and the motion vector compression / decompression unit 209 of the video decoding device performs the same function as the motion vector compression / expansion unit 112 of the video encoding device.

  Each flag of the bit stream decoded by the variable length decoding unit 201 will be described. It is determined from the flags described in the SPS header of the bitstream whether or not to reduce the memory usage by applying a motion vector selection method to the entire bitstream. The flag motion vector buffer comp flag is determined from the SPS header. If the motion vector buffer comp flag is true (1), memory usage is reduced. If the motion vector buffer comp flag is false (0), it is not implemented. . Further, when the motion vector buffer comp flag is true (1), the flag motion vector buffer comp ratio is read from the SPS header and input to the predetermined range setting unit 301. The predetermined range setting unit 301 calculates the number MVBufferCompRatio of the predetermined range in the vertical and horizontal directions based on the above formula. Every time the SPS header is decoded, MVBufferCompRatio is recalculated. Until then, the MVBufferCompRatio calculated before that is recorded in the predetermined range setting unit 301 and used, so this processing is performed for each picture. There is no need to do.

  Thus, the motion vector selection method is performed based on the flag decoded from the bitstream. The motion vector compression / decompression operation between the memory 207 and the motion vector compression / decompression unit 209 is the same as that of the memory 110 and the motion vector compression / decompression unit 112 in the video encoding apparatus, and thus description thereof is omitted.

As described above, in the video encoding device and video decoding device including the motion vector selection method according to the present invention, the motion vectors recorded in the memories 110 and 207 are stored in the motion vector compression / decompression units 112 and 209. Reduction of memory usage of motion vectors is realized by performing compression processing and re-recording. The actual motion vector compression processing is performed by the motion vector compression unit 303 in the motion vector compression / expansion units 112 and 209, and an optimum representative value is selected from a plurality of motion vectors in a predetermined range. Below, the specific Example regarding representative value selection is shown.
[First Example]
The motion vector selection method according to the present invention will be described with reference to a first embodiment of motion vector compression. Here, description will be made by defining each block size under the following conditions. Note that this condition is also defined and described in the first and subsequent examples.

    -The minimum coding block size is 8 x 8 pixels.

    ・ The size of the maximum coding block is 32 × 32 pixels.

・ Set the number of MVBufferCompRatio in the specified range to 4.
That is, the motion vector buffer comp ratio is 2.

    The size of the minimum prediction block is 4 × 4 pixels, which is a size obtained by dividing the minimum coding block into four.

    The memory area corresponding to the smallest prediction block belongs to the prediction block as auxiliary information of the prediction block, in addition to the motion information such as the motion vector, reference picture number, prediction mode, and prediction direction, before compression of the motion vector. Information on the encoded block is temporarily recorded. Information (depth) indicating the number of times of division from the maximum coding block to be the size of the coding block to which the prediction block belongs, information partMode representing the coding block division mode, and the like.

  FIG. 15 shows an example of attached information of the prediction block recorded in the memory area corresponding to the minimum prediction block. A block surrounded by a dotted line in FIG. 15 is a minimum coding block, and a thick solid line represents division. FIG. 15A shows a case where there is no division. At this time, depth = 0 and partMode = 2N × 2N are recorded in all the minimum prediction blocks. Fig. 15 (b) shows the case of dividing into four parts with respect to Fig. 15 (a), and there is a difference in motion information in the rectangle divided into four, but since the division was performed once, all the minimum In the prediction block, depth = 1 and partMode = 2N × 2N are recorded. FIG. 15 (c) shows a case where the upper right block is further divided into four with respect to FIG. 15 (b). Since there is no division processing of the upper left, lower left and lower right blocks, the depth and partMode in FIG. 15B are recorded in each minimum prediction block. In each minimum prediction block of the upper right block divided into four, depth = 2 and partMode = 2N × 2N are recorded. FIGS. 15 (d) and 15 (e) show a case where the image is divided into two in the vertical direction or the horizontal direction with respect to FIG. At this time, the depth is not added and the division mode is changed. That is, depth = 0 and partMode = N × 2N in the case of FIG. 15D, and depth = 0 and partMode = 2N × N in the case of FIG. 15E are recorded in all the minimum prediction blocks. Thus, when the coding block is divided into four, depth is added to the minimized prediction block belonging to the divided lower coding block, and when divided into two in the horizontal or vertical direction, the divided lower block The division mode of the minimized prediction block belonging to the encoded block is changed to represent the state of the minimum prediction block.

  In the first embodiment according to the present invention, a prediction block having the largest area in a predetermined range is selected, and a motion vector of the prediction block is selected as a representative value. By storing the representative motion vector as a predicted motion vector candidate used in the prediction of the motion vector of the subsequent image, the storage capacity can be reduced as compared with the case of storing the motion vectors of all the prediction blocks. I can do it. The same motion information is recorded in the memory of the smallest prediction block that constitutes the prediction block, and selecting the prediction block having the largest area within the predetermined range selects a motion vector having a large number within the predetermined range. This is equivalent to doing so and minimizes the motion error within a predetermined range. By using this motion vector as a motion vector predictor candidate for use in the prediction of the motion vector of the following picture, the prediction accuracy is improved. Coding efficiency can be improved. In addition, as a condition for selecting a representative value of motion vectors, when the prediction blocks occupying within a predetermined range are the same, the earlier one is preferentially selected in the processing order. Further, the minimum prediction block for selecting the representative value of the motion vector is the upper left block among the prediction blocks selected within a predetermined range. Since the smallest prediction block at the upper left in the prediction block is the head of the prediction block, there is an advantage that the address conversion to the memory is easy and the accessibility is high.

  FIG. 16 is a flowchart for explaining the operation of the motion vector compressing unit 303 in the motion vector compressing / decompressing units 112 and 209 implementing the first embodiment according to the present invention. First, the memory area in which the motion information of the predetermined range specified by the predetermined range position setting unit 302 is recorded is accessed from the memories 110 and 207, and the motion information of the minimum prediction block within the predetermined range is received by the motion vector compression unit 303. Input (S101). As shown in FIG. 17, input motion information is motion information of MVBufferCompRatio × MVBufferCompRatio minimum prediction blocks. In the first embodiment, the predetermined range is composed of 16 minimum prediction blocks, and is a rectangular area of 16 × 16 pixels. This is the same size as the divided encoded block when the maximum encoded block is divided once. Next, motion information of the upper left minimum prediction block shown in FIG. 17 is designated from the input motion information in a predetermined range, and depth, partMode, and prediction mode are read as information used for determination (S102). First, a determination process is performed for depth, and it is determined whether depth is 0 (S103). When the depth is 0, this corresponds to no division in the maximum encoded block, and the maximum encoded block becomes a prediction block, so the maximum encoded block has one piece of motion information. For this reason, as shown in FIG. 18 (a), the minimum prediction blocks in the four predetermined ranges existing in the maximum coding block are all configured with the same motion information. Next, it is determined whether the prediction mode of the upper left minimum prediction block is intra (S106). When the prediction mode is intra, the other minimum prediction blocks within the predetermined range are all intra, and no motion vector exists, so (0, 0) is output as a motion vector (S107). If the prediction mode is not intra, the motion vector of the minimum prediction block at the upper left of the predetermined range is selected as a representative value, and the process ends (S108). On the other hand, if the depth is not 0, the process proceeds to determination of whether the depth is 1 (S104). If the depth is not 1, the inside of the predetermined range is divided into four or more divisions. When the depth is 1, the maximum coding block is divided into four, so as shown in FIG. 18 (b), it is divided into four equal coding blocks. Will match. For this reason, when partMode is 2N × 2N in the next determination of partMode (S105), this is equivalent to the case where the inside of the coding block is not divided into two, and this coding block, that is, the smallest prediction block within the predetermined range is all the same. It consists of motion information. Since the state of this predetermined range is the same as that when depth is 0, the process proceeds to the processing after determination (S106) whether or not the prediction mode of the upper left minimum prediction block is intra. On the other hand, when the partMode is not 2N × 2N, that is, there is a division in the horizontal or vertical direction inside the coding block, the process proceeds to the 2-division selection determination and ends.

  Next, the two-division selection determination will be described using the flowchart of FIG. The depth is 1 and the partMode is N × 2N or 2N × N when the predetermined range is divided into two in the horizontal or vertical direction, which corresponds to FIGS. 18 (c) and 18 (d). In this case, the divided left and right or upper and lower prediction blocks cannot be determined because they occupy the same area. Here, according to the above-described conditions for selecting the representative value, the earlier one is preferentially selected as the processing order. Of the prediction blocks divided within the predetermined range, the prediction mode of the minimum prediction block at the upper left of the left or upper prediction block preceding in the processing order is read (S201). It is determined whether or not the prediction mode is intra (S202). If the prediction mode is not intra, the motion vector of the minimum prediction block is output and the process ends (S204). When the prediction mode is intra, the prediction mode of the minimum prediction block at the upper left of the other right or lower prediction block is read (S203). It is determined whether or not the prediction mode is intra (S205). When the prediction mode is not intra, the motion vector of the minimum prediction block is output and the process ends (S207). If the prediction mode is intra, (0, 0) is output as the motion vector, and the process ends (S206).

Next, the 4-division selection determination will be described. This situation occurs when the depth is 2, which is equally divided into four within a predetermined range, and four minimum coding blocks are arranged as shown in FIG. 18 (e). Here, in order to facilitate the explanation, the names A, B, C, and D are assigned to the minimum coding blocks shown in FIG. 18 (e) in the order of processing from the upper left to the lower right. Each minimum encoded block is divided into four types of predicted blocks as shown in FIG. A solid line in the minimum coding block in FIG. 20 represents division, a dotted line represents a boundary of the minimum prediction block, and a number is assigned to each of the four minimum prediction blocks. According to the representative value selection conditions described above, the minimum prediction block at the upper left (the minimum prediction block of “00” in FIG. 20) is selected regardless of the four division patterns in FIG. The division states in the four minimum encoded blocks indicated by 18 (e) are compared, and the minimum prediction block on the upper left of the minimum encoded block with the smallest number of prediction blocks generated by division may be selected. If the number of divided prediction blocks is small, the number of minimum prediction blocks having the same motion information increases, and the area that occupies the predetermined range increases. FIG. 21 shows an example of a divided arrangement of four minimum coding blocks within a predetermined range. In the case of FIG. 21 (a), C is selected because the minimum coding block other than C is divided. In the case of FIG. 21 (b), since all the minimum encoded blocks are divided into four, A which is first in the processing order is selected. In the case of FIG. 21 (c), A and B are divided into four, and C and D are divided into two. Therefore, the first C in the processing order is selected. In the case of FIG. 21 (d), since A and B are not divided and C and D are divided, A is selected in the processing order. However, when the prediction mode of A is intra, the processing order is as follows. B is selected. A motion vector is output as a representative value from the minimum prediction block (gray rectangular area in FIG. 21) at the upper left of the selected minimum coding block. The flow of such a 4-division selection determination process is summarized in the flowchart of FIG. First, a register for temporarily storing an optimum value is initialized. The register is a MaxPart that records the division mode with the smallest number of divisions in the division state of the encoded block, and an MvReg that records the motion vector of the minimum prediction block in the upper left of the encoded block at that time, each MaxPart = 4, MvReg = (0,0) is initialized (S301). Here, values assigned to the division mode will be described. FIG. 23 shows an example of the definition of the block division mode in the syntax of the bit stream of the moving image encoded by the moving image encoding apparatus according to the present invention. Inter partioning idc is defined as an index value for partMode indicating the division mode, and this index value is encoded. The priority in the split mode in the representative selection of motion vectors is
2N × 2N> 2N × N or N × 2N> N × N
However, the index value assignment is reversed. In the first embodiment, since the definition of the division mode in FIG. 23 is followed, the smaller one of the comparisons by the division mode between the minimum coding blocks is prioritized. Therefore, “4”, which is larger than the index value of N × N, is set as the initial value of MaxPart. Next, a counter is set to A as the minimum coding block to be subjected to the determination process first (S302). The motion information of the minimum prediction block at the upper left of the minimum coding block set in the counter is read (S303). It is determined whether or not the prediction mode is intra (S304). When the prediction mode is intra, the determination process is not performed, and the process proceeds to the counter determination (S307). If the prediction mode is not intra, partMode and MaxPart are compared (S305). When partMode is smaller than MaxPart, that is, when the size of the prediction block including the minimum coding block set in the currently processed counter is large, MaxPart is updated to partMode, and the motion vector at that time is stored in MvReg Is done. If partMode is equal to or greater than MaxPart, the process proceeds to counter determination (S307). In the determination of the counter, it is determined whether or not the determination process has been completed for the four minimum encoded blocks (S307). If the counter is D, it corresponds to the last block among the four, so the determination process is terminated and the motion vector stored in MvReg is output as a representative value (S309). If the counter is not D, the counter is updated (S308). Update in the order of A, B, C, D, and repeat the processing from S303.

As described above, representative values of motion vectors in a predetermined range are selected. The selected representative value is sent to the memory 110 and the video decoding device memory 207 of the video encoding device, and is recorded in a recording area corresponding to a position in a predetermined range.
[Second Example]
A second embodiment of motion vector compression will be described with respect to a motion vector selection method according to the present invention. The basic selection method follows the first embodiment, and changes the priority condition of the minimum prediction block having a motion vector to be selected as a representative value in the 4-division selection determination within a predetermined range. In the first embodiment, the minimum prediction block for selecting the representative value of the motion vector is the upper left block of the prediction blocks selected within the predetermined range. However, in the second embodiment, the block near the center in the predetermined range is selected. Select the smallest prediction block. For example, when the predetermined range of the current picture is divided into four minimum encoded blocks within the predetermined range as shown in FIG. 21B, the motion vectors of the individual minimum encoded blocks are different. There are many cases. When coding the next picture, if the coding block at the same position is D, the motion vector at the same position of the reference picture of D is a motion within a predetermined range according to the selection method of the first embodiment. The upper left of A is selected as the representative value of the vector. The reference vector for D is the motion vector of the block located at the same position as D in the reference picture, but the motion vector of the upper left prediction block of A is referenced as a candidate for the predicted motion vector by compressing the motion vector. Therefore, the prediction accuracy may be reduced. Therefore, in the second embodiment, in order to further improve the accuracy of the motion vector predictor, the minimum predicted block closer to the center within the predetermined range is preferentially selected. In the case of the above example, the smallest prediction block at the lower right of A is selected. Since the minimum prediction block at the lower right of A is closer to D than the upper left of A, it becomes a predicted motion vector candidate close to the motion vector of D, and an improvement in prediction accuracy can be expected. In particular, such a selection is effective for encoding a coding block having a predetermined range or less.

  An example of specific motion vector selection in the second embodiment will be described with reference to FIG. FIG. 24 is a determination example corresponding to the four-division selection determination example of the first embodiment. In the case of FIG. 24 (a), C is selected. Since the motion vectors of the minimum prediction block in C are the same, the motion vector of the minimum prediction block in the upper left is selected. In the case of FIG. 24 (b), A is selected, and the motion vector of the minimum prediction block at the lower right of A near the center of the predetermined range is selected. In the case of FIG. 24 (c), C is selected. Since C is divided into two in the vertical direction, since the right prediction block is closer to the center of the predetermined range, the motion vector of the minimum prediction block above the right prediction block is selected. In the case of FIG. 24 (d), B is selected because the prediction mode of A is intra. Since the motion vectors of the minimum prediction block in B are the same, the motion vector of the minimum prediction block at the upper left is selected.

As described above, in the four-division selection determination within the predetermined range, the minimum prediction block for selecting the representative value of the motion vector is the minimum prediction block closer to the center within the predetermined range among the prediction blocks selected within the predetermined range. Selected. When actually performing the 4-division selection determination, the minimum prediction block for selecting the representative value of the motion vector is selected from the pattern of the partMode shown in FIG. 25 and the block selected within the predetermined range. The numbers in FIG. 25 are numbers assigned to the minimum prediction block in the encoded block shown in FIG. 20, and the motion vector of the minimum prediction block corresponding to this number is selected.
[Third embodiment]
A third embodiment of motion vector compression will be described with respect to a motion vector selection method according to the present invention. The basic selection method follows the first embodiment, and changes the priority condition of the minimum prediction block having a motion vector to be selected as a representative value in the 4-division selection determination within a predetermined range. In the third embodiment, based on the motion information of the four smallest predicted blocks in the center within the predetermined range shown in gray in FIG. 26, representative values of motion vectors are selected from the four. Here, in order to facilitate the following description, the numbering shown in FIG. 26 is given to the minimum prediction block arranged from the upper left to the lower right arranged in the center within the predetermined range.

  The process flow of the 4-division selection determination of the third embodiment will be described with reference to the flowchart of FIG. First, a register for temporarily storing an optimum value is initialized. Registers are MinDepth, which records the smallest number of divisions in the division state of the smallest prediction block, MaxPart, which records the division mode with the smallest division number in the division state of the smallest prediction block, and the coding block at that time It is MvReg which records the motion vector of the minimum prediction block at the upper left, and is initialized to MinDepth = 4, MaxPart = 4, and MvReg = (0, 0), respectively (S401). Since the initial value of MinDepth cannot be divided into four smaller than the minimum prediction block, the number of divisions until the maximum encoding block is divided into the minimum prediction block is set. Here, since the maximum coding block is 32 × 32 pixels and the minimum prediction block is 4 × 4 pixels, the maximum coding block is less than 4 × 4 pixels in 4 divisions, so “4” is set. Is done. The initialization of MaxPart and MvReg follows the setting of the first embodiment described above. Next, the counter count is set to 0 as the minimum coding block to be subjected to the determination process first (S402). The motion information of the minimum prediction block set in the counter is read (S403). It is determined whether or not the prediction mode is intra (S404). When the prediction mode is intra, the determination process is not performed, and the process proceeds to the counter determination (S408). When the prediction mode is not intra, the depth is compared with MinDepth (S405). When the depth is smaller than MinDepth, that is, when the number of prediction blocks including the smallest prediction block set in the currently processed counter is small, it means that the area occupied by the prediction block is large within the specified range. To do. If the depth is equal to or greater than MinDepth, the process proceeds to counter determination (S408). If the depth is smaller than MinDepth, partMode and MaxPart are compared (S406). When partMode is smaller than MaxPart, that is, when the size of the prediction block including the minimum prediction block set in the currently processed counter is large, which means that the area occupied by the prediction block is large within a predetermined range. . If partMode is equal to or greater than MaxPart, the process proceeds to counter determination (S408). If partMode is smaller than MaxPart, MinDepth is updated to depth and MaxPart is updated to partMode, and the motion vector of the minimum prediction block at that time is stored in MvReg. In the determination of the counter, it is determined whether or not the determination process has been completed for the four minimum encoded blocks (S408). If count = 3, the determination process is terminated, and the motion vector stored in MvReg is output as a representative value (S410). If count ≠ 3, the count is incremented by 1, and the processing after S403 is repeated (S409).

As described above, representative values of motion vectors in a predetermined range are selected. The selected representative value is sent to the memory 110 and the video decoding device memory 207 of the video encoding device, and is recorded in a recording area corresponding to a position in a predetermined range.
[Fourth embodiment]
A fourth embodiment of motion vector compression will be described with respect to a motion vector selection method according to the present invention. The basic selection method follows the first embodiment, but differs from the first embodiment in that a representative value of a motion vector is selected from the prediction block based on the size of the prediction block to be divided. Even if a motion vector of a prediction block smaller than the predetermined range is selected, it cannot be said that it is a highly reliable representative value and it cannot be said that the prediction accuracy is improved as a prediction motion vector candidate. Restrict.

  The divided predetermined range shown in FIG. 28 will be described as an example. The solid line in the predetermined range in FIG. 28 indicates the boundary by four divisions, and the dotted line indicates the boundary of the minimum prediction block. Names are given in order from the upper left to the lower right for the minimum coding block in which the predetermined range is divided into four. It shall be represented by A, B, C, D. The predetermined range in FIG. 28 is divided into four and divided into four minimum encoded blocks. The state of each minimum coding block is divided into four until A and C become further minimum prediction blocks, B and D are divided into two, and the division modes are represented by 2N × N and N × 2N, respectively. Depth = 3 and partMode = 2N × 2N are set in the smallest prediction block in A and C, depth = 2, partMode = 2N × N in B, depth = 2, partMode = N × 2N in C Is set. When the motion vector selection according to the first embodiment is applied to the predetermined range in FIG. 28, the left smallest prediction block in the upper two prediction blocks divided into B shown by the gray rectangle in FIG. 28 is selected. Then, the motion vector is selected. The right minimum prediction block in the upper prediction block divided into two bisects also has a selected motion vector, but only two of the 16 minimum prediction blocks in a given range, the prediction motion vector The accuracy of the candidate is low. In the fourth embodiment, a prediction block that is less than the minimum determination range set based on the size of a predetermined range is limited so as not to be selected as a motion vector. Further, when a predicted block that is equal to or greater than the minimum determination range is found within the predetermined range, the determination processing time is improved by forcibly stopping the subsequent determination processing.

  FIG. 29 shows a flowchart of the determination process of the fourth embodiment. As a limiting condition in the fourth embodiment, an area in which the number of vertical and horizontal pixels in a predetermined range is halved is set as the minimum determination range. The minimum determination range is 8 × 8 pixels, the depth that matches this condition is 2, and the partMode is 2N × 2N, and the determination process is performed for the prediction block that includes the minimum prediction block with motion information that is less than this condition do not do. First, the memory area in which the motion information of the predetermined range specified by the predetermined range position setting unit 302 is recorded is accessed from the memories 110 and 207, and the motion information of the minimum prediction block within the predetermined range is input to the motion vector compression unit 303 (S501). Next, a register MvReg for temporarily recording a motion vector is initialized. The register MvReg is initialized to (0, 0) (S502). The counter is set to A as the minimum coding block for performing the determination process (S503). In the fourth embodiment, since the minimum determination range is the same size as the minimum encoded block, the process is performed in units of the minimum encoded block within the predetermined range. The motion information of the minimum prediction block of the minimum coding block set in the counter is read (S504). If there is no division in the minimum coding block, the four minimum prediction blocks in the minimum coding block all have the same motion information, so the motion information of any minimum prediction block may be read. The minimum prediction block at the upper left is specified. First, it is determined whether or not the prediction mode is intra (S505). When the prediction mode is intra, the determination process is not performed, and the process proceeds to the counter determination (S510). If the prediction mode is not intra, a determination is made regarding the depth of the motion information of the designated minimum prediction block (S506). When the depth is smaller than 2, the minimum coding block is included in the prediction block larger than the minimum determination range, so the determination is stopped and the process proceeds to the motion vector update (S509). When the depth is 2 or more, it is determined whether or not the depth is equal to 2 (S507). If the depth is equal to 2, proceed to split mode determination. When the depth is not 2, that is, when the depth is 3 or more, there are 4 divisions in the minimum coding block set in the counter, and there is a prediction block smaller than the minimum judgment range. The determination is stopped and the process proceeds to the counter determination (S510). In the determination of the division mode, it is determined whether or not partMode is equal to 2N × 2N (S508). When partMode is 2N × 2N, since the depth is 2, the minimum encoded block has the same size as the minimum determination range, so the determination is stopped and the process proceeds to motion vector update (S509). If partMode is not 2N × 2N, there are two horizontal or vertical divisions within the minimum coding block set in the counter, and there will be a prediction block smaller than the minimum decision range. The determination is stopped and the process proceeds to the counter determination (S510). In the update of the motion vector, after the motion vector of the designated minimum prediction block is stored in MvReg (S509), the motion vector stored in MvReg is output as a representative value within a predetermined range (S512), and the process ends. On the other hand, in the determination of the counter, it is determined whether the counter is D (S510). If the counter is D, the determination process has been completed for all the minimum coding blocks within the predetermined range, so the process proceeds to S512, and the motion vector stored in MvReg is output as a representative value of the predetermined range ( S512) and finish. If the counter is not D, the counter is updated (S511). Update in the order of A, B, C, D, and repeat the processing from S503.

  As described above, representative values of motion vectors in a predetermined range are selected. The selected representative value is sent to the memory 110 and the video decoding device memory 207 of the video encoding device, and is recorded in a recording area corresponding to a position in a predetermined range.

  In the fourth embodiment, when there is only a prediction block less than the minimum determination range within the predetermined range, (0, 0) set as the initial value of the register MvReg is output as the representative value of the motion vector in the predetermined range. However, the motion vector of the first minimum encoded block may be set as an initial value. Also, when determining in order for the smallest encoded block within the predetermined range, the determination process is stopped when a prediction block corresponding to the minimum determination range is first found. Instead, the determination process may be changed so that the motion vector at that time is temporarily stored in MvReg, and the motion vector of the prediction block corresponding to the minimum determination range is finally output within the predetermined range. .

  The above processing relating to encoding and decoding can be realized as a transmission, storage, and reception device using hardware, and is stored in a ROM (Read Only Memory), a flash memory, or the like. It can also be realized by firmware or software such as a computer. The firmware program and software program can be recorded on a computer-readable recording medium, provided from a server through a wired or wireless network, or provided as a data broadcast of terrestrial or satellite digital broadcasting Is also possible.

101 Motion vector detector
102 Subtraction part
103 Orthogonal transformation / quantization unit
104 Variable length encoder
105 Inverse quantization and inverse orthogonal transform
106 Motion compensation unit
107 Weighted predictor
108 Adder
109 Deblocking filter section
110 memory
111 Predictive motion vector selector
112 Motion vector compression / decompression unit
201 Variable length decoder
202 Inverse quantization and inverse orthogonal transform
203 Motion compensation unit
204 Weighted prediction part
205 Adder
206 Deblocking filter section
207 memory
208 Predictive motion vector selector
209 Motion vector compression / decompression unit
301 Predetermined range setting section
302 Predetermined range position setting section
303 Motion vector compression unit
304 Motion vector stretcher
305 switch
306 switch
307 Temporary memory

Claims (10)

A video decoding device that decodes a bitstream in which the video is encoded using a motion vector in units of blocks obtained by dividing each picture of the video,
Using motion information including a motion vector decoded in units of blocks within a predetermined range from the bitstream, and using a parameter indicating the size of a prediction block to which the block within the predetermined range decoded from the bitstream belongs Determining a prediction block having a large area occupying the predetermined range, storing the determined motion vector of the prediction block in a memory as a representative value of the predetermined range, and at the time of decoding the next picture, A motion vector compression / decompression unit that selects a representative value of the motion vector from the memory according to the position of the prediction block;
A predicted motion vector selection unit that decodes a predicted motion vector using a representative value of the motion vector selected by the motion vector compression / decompression unit as a candidate for a predicted motion vector;
A moving picture decoding apparatus comprising: The video decoding device according to claim 1,
The prediction block is a rectangular pixel area having one or a plurality of the divided blocks, and is divided so as to have the same motion information and a parameter representing the size of the prediction block. Decryption device. The video decoding device according to claim 1,
In the divided block, a reference picture number and a prediction mode are stored as the motion information in addition to a motion vector, and the parameter indicating the size of the prediction block to which the divided block belongs is set as the size of the prediction block. A moving picture decoding apparatus, wherein a division mode indicating the number of divisions up to and a shape of the prediction block is stored. The video decoding device according to claim 1,
The predetermined range is defined by a square rectangular array that equally divides the motion vector data array arranged two-dimensionally so as to correspond to the picture in the memory, and the number of blocks divided as the side length of the square rectangular array A video decoding device characterized in that it obtains a parameter representing the above from a bit stream. The video decoding device according to claim 1,
The motion vector compression / decompression unit selects a motion vector of the prediction block as a representative value by prioritizing the previous prediction block in decoding order when a plurality of prediction blocks remain as selection candidates in the predetermined range. A moving picture decoding apparatus characterized by: The video decoding device according to claim 1,
The motion vector compression / decompression unit represents a motion vector of the prediction block by giving priority to a prediction block located near the center of the predetermined range when a plurality of prediction blocks remain as selection candidates in the predetermined range. A moving picture decoding apparatus characterized by selecting as a value. The video decoding device according to claim 1,
The motion vector compression / decompression unit has the largest predicted block size to which the plurality of blocks belong based on motion information of a plurality of blocks arranged in the center of the predetermined range and a parameter representing the size of the predicted block. A moving picture decoding apparatus, wherein a motion vector of a block is selected as a representative value. The video decoding device according to claim 1,
The motion vector compression / decompression unit previously determines a minimum determination range for performing determination of motion vector selection based on the size of the predetermined range, and a prediction block in the predetermined range is larger than the minimum determination range. In some cases, the motion decoding apparatus is characterized in that the motion vector of the prediction block is selected as a representative value, and the determination process is terminated. A moving picture decoding method for decoding a bitstream in which the moving picture is encoded using a motion vector in units of blocks obtained by dividing each picture of the moving picture,
Using motion information including a motion vector decoded in units of blocks within a predetermined range from the bitstream, and using a parameter indicating the size of a prediction block to which the block within the predetermined range decoded from the bitstream belongs Determining a prediction block having a large area occupying the predetermined range, storing the determined motion vector of the prediction block in a memory as a representative value of the predetermined range, and at the time of decoding the next picture, And a motion vector compression / decompression step for selecting a representative value of the motion vector from the memory according to the position of the prediction block;
A predicted motion vector selection step of decoding a predicted motion vector using a representative value of the motion vector selected in the motion vector compression / decompression step as a predicted motion vector candidate;
A moving picture decoding method comprising: A moving picture decoding program for causing a computer to execute an operation of decoding a bitstream in which the moving picture is encoded using a motion vector in units of blocks obtained by dividing each picture of the moving picture,
Using motion information including a motion vector decoded in units of blocks within a predetermined range from the bitstream, and using a parameter indicating the size of a prediction block to which the block within the predetermined range decoded from the bitstream belongs Determining a prediction block having a large area occupying the predetermined range, storing the determined motion vector of the prediction block in a memory as a representative value of the predetermined range, and at the time of decoding the next picture, And a motion vector compression / decompression step for selecting a representative value of the motion vector from the memory according to the position of the prediction block;
A predicted motion vector selection step of decoding a predicted motion vector using a representative value of the motion vector selected in the motion vector compression / decompression step as a predicted motion vector candidate;
Is a moving picture decoding program for causing a computer to execute.

Images