JP2020145494A - Image coding device, image coding method, and image coding program - Google Patents

Abstract

To provide a technique for improving coding efficiency by performing block division suitable for image coding and decoding.SOLUTION: An image coding device includes a merge candidate construction unit that builds a merge candidate list that includes spatial merge candidates, a merge candidate selection unit that selects one selected merge candidate from the merge candidate list on the basis of a merge index decoded from a coded stream, a bi-prediction conversion unit that converts the selected merge candidate into bidirectional prediction motion information consisting of L0 prediction and L1 prediction when the selected merge candidate is motion information of uni-prediction, and a motion information allocation unit that divides a coded block into 4x4 sub-blocks, and allocates any one of motion information of the L0 prediction of the bi-prediction, motion information of the L1 prediction of the bi-prediction, and motion information on both the L0 prediction and the L1 prediction of the bi-prediction for each of the sub-blocks.SELECTED DRAWING: Figure 1

Description

The present invention relates to an image coding and decoding technique for dividing an image into blocks and performing prediction.

In image coding and decoding, an image to be processed is divided into blocks which are a set of a predetermined number of pixels, and processing is performed in block units. Coding efficiency is improved by dividing into appropriate blocks and appropriately setting in-screen prediction (intra prediction) and inter-screen prediction (inter prediction).

In the coding / decoding of moving images, the coding efficiency is improved by inter-prediction that predicts from the coded / decoded pictures. Patent Document 1 describes a technique for applying an affine transformation at the time of inter-prediction. In moving images, it is not uncommon for an object to undergo deformation such as enlargement / reduction and rotation, and efficient coding becomes possible by applying the technique of Patent Document 1.

Japanese Unexamined Patent Publication No. 9-172644

However, since the technique of Patent Document 1 involves conversion of an image, there is a problem that the processing load is large. In view of the above problems, the present invention provides a low-load and efficient coding technique.

In one aspect of the present invention that solves the above problems, a merge candidate construction unit that constructs a merge candidate list including spatial merge candidates, and one selected merge candidate from the merge candidate list based on the merge index decoded from the coded stream. A merge candidate selection unit that selects a merge candidate, and a bi-prediction conversion unit that converts the selection merge candidate into bidirectional prediction motion information consisting of L0 prediction and L1 prediction when the selection merge candidate is simple prediction movement information. The coded block is divided into 4x4 sub-blocks, and for each sub-block, the movement information of the L0 prediction of the twin prediction, the movement information of the L1 prediction of the double prediction, and the movement of both the L0 prediction and the L1 prediction of the double prediction. It has a motion information allocation unit that allocates any of the information.

According to the present invention, highly efficient image coding / decoding processing can be realized with a low load.

It is a block diagram of the image coding apparatus which concerns on embodiment of this invention. It is a block diagram of the image decoding apparatus which concerns on embodiment of this invention. It is a flowchart for demonstrating the operation which divides a tree block. It is a figure which shows a mode that the input image is divided into a tree block. It is a figure explaining the z-scan. It is a figure which shows the division shape of a block. It is a flowchart for demonstrating the operation which divides a block into four. It is a flowchart for demonstrating the operation which divides a block into two or three. This is a syntax for expressing the shape of block division. It is a figure for demonstrating an intra prediction. It is a figure for demonstrating the reference block of an inter-prediction. This is a syntax for expressing the coded block prediction mode. It is a figure which shows the correspondence of a syntax element and a mode about inter prediction. It is a figure for demonstrating the affine transformation motion compensation of two control points. It is a figure for demonstrating the affine transformation motion compensation of three control points. It is a block diagram of the detailed structure of the inter-prediction unit 102 of FIG. It is a block diagram of the detailed structure of the normal prediction motion vector mode derivation part 301 of FIG. It is a block diagram of the detailed structure of the normal merge mode derivation part 302 of FIG. It is a flowchart for demonstrating the normal prediction motion vector mode derivation process of the normal prediction motion vector mode derivation unit 301 of FIG. It is a flowchart which shows the processing procedure of the normal prediction motion vector mode derivation processing. It is a flowchart explaining the procedure of a normal merge mode derivation process. It is a block diagram of the detailed structure of the inter-prediction unit 203 of FIG. It is a block diagram of the detailed structure of the normal prediction motion vector mode derivation unit 401 of FIG. It is a block diagram of the detailed structure of the normal merge mode derivation part 402 of FIG. It is a flowchart for demonstrating the normal prediction motion vector mode derivation process of the normal prediction motion vector mode derivation unit 401 of FIG. It is a block diagram of the detailed structure of the sub-block prediction motion vector mode derivation unit 303 of FIG. It is a block diagram of the detailed structure of the sub-block prediction motion vector mode derivation unit 403 of FIG. It is a block diagram of the detailed structure of the sub-block merge mode derivation unit 304 of FIG. It is a block diagram of the detailed structure of the sub-block merge mode derivation unit 404 of FIG. 22. It is a figure explaining the affine inheritance prediction motion vector candidate derivation. It is a figure explaining the affine construction prediction motion vector candidate derivation. It is a figure explaining the affine inheritance merge candidate derivation. It is a figure explaining the affine construction merge candidate derivation. It is a flowchart of affine inheritance prediction motion vector candidate derivation. It is a flowchart of affine construction prediction motion vector candidate derivation. It is a flowchart of affine inheritance merge candidate derivation. It is a flowchart of affine construction merge candidate derivation. It is a flowchart explaining the history prediction motion vector candidate list initialization / update processing procedure. It is a flowchart of the same element confirmation processing procedure in the history prediction motion vector candidate derivation processing procedure. It is a flowchart of the element shift processing procedure in the history prediction motion vector candidate derivation processing procedure. It is a flowchart explaining the history prediction motion vector candidate derivation processing procedure. It is a flowchart explaining the history merge candidate derivation processing procedure. It is a figure explaining an example of the history prediction motion vector candidate list update processing. It is a flowchart for demonstrating the operation of the sub-block time merge candidate derivation unit 381. It is a flowchart for demonstrating the process of deriving the adjacent movement information of a block. It is a flowchart for demonstrating the process of deriving a temporal motion vector. It is a flowchart for demonstrating the derivation of inter prediction information. It is a flowchart for demonstrating the process of deriving the sub-block motion information. It is a figure for demonstrating the temporal context of a picture. It is a flowchart for demonstrating the derivation process of the time prediction motion vector candidate in a normal prediction motion vector mode derivation unit 301. It is a flowchart for demonstrating the derivation process of ColPic in the derivation process of the time prediction motion vector candidate in a normal prediction motion vector mode derivation part 301. It is a flowchart for demonstrating the process of deriving the coded information of ColPic in the process of deriving the time-predicted motion vector candidate in a normal prediction motion vector mode derivation unit 301. It is a flowchart for demonstrating the derivation process of inter-prediction information. It is a flowchart which shows the derivation processing procedure of the inter-prediction information of a coded block when the inter-prediction mode of the coded block colCb is bi-prediction (Pred_BI). It is a flowchart for demonstrating the scaling operation processing procedure of a motion vector. It is a flowchart for demonstrating the derivation process of a time merge candidate. It is a figure for demonstrating the motion compensation prediction at the time of L0 prediction, and the reference picture (RefL0Pic) of L0 is at the time before the processing target picture (CurPic). It is a figure for demonstrating motion compensation prediction at the time of L0 prediction, and the reference picture of L0 prediction is at the time after the processing target picture. It is a figure for clarifying the motion compensation prediction at the time of the bi-prediction, when the reference picture of L0 prediction is at the time before the processing target picture, and the reference picture of L1 prediction is at the time after the processing target picture. .. It is a figure for demonstrating the prediction direction of the motion compensation prediction at the time when the reference picture of L0 prediction and the reference picture of L1 prediction are at the time before the picture to be processed in the bi-prediction. It is a figure for demonstrating the prediction direction of the motion compensation prediction at the time when the reference picture of L0 prediction and the reference picture of L1 prediction are at the time after the processing target picture in the bi-prediction. It is a flowchart explaining the average merge candidate derivation processing procedure. It is a table which shows the information about a merge difference motion vector. It is a figure explaining the derivation of a merge difference motion vector. It is a figure explaining the prediction of the triangular merge mode. It is a figure explaining the derivation of the triangular merge candidate list. It is a figure explaining the acquisition of motion information from a triangular merge candidate list. It is a figure explaining the process of acquiring motion information from a triangular merge candidate list, and calculating a code amount and a strain amount. This is a syntax for expressing the merge mode. It is a figure explaining the acquisition of motion information from a triangular merge candidate list. It is a figure explaining the acquisition of the movement information from the merge candidate list.

The technology and technical terms used in this embodiment are defined.

<Tree block>
In the embodiment, the image to be encoded / decoded is evenly divided into a predetermined size. This unit is defined as a tree block. As shown in FIG. 4, in the present embodiment, the size of the tree block is set to 128 × 128 pixels, but the size of the tree block is not limited to this, and any size may be set. The tree blocks of the processing target (corresponding to the coding target in the coding process and the decoding target in the decoding process) are switched in the raster scan order, that is, in the order of left to right and top to bottom. The inside of each tree block can be further recursively divided. The block to be coded / decoded after the tree block division is defined as the coded block. In addition, tree blocks and coded blocks are collectively defined as blocks. Efficient coding is possible by performing appropriate block division. The size of the tree block may be a fixed value previously agreed between the coding device and the decoding device, or the size of the tree block determined by the coding device may be transmitted to the decoding device.

<Prediction mode>
Processed image to be processed in units of coded blocks to be processed (in the coding process, the coded signal is used for the decoded image, image signal, tree block, block, coded block, etc., and in the decoding process Intra-prediction (MODE_INTRA) that predicts from the image signal around the image, image signal, tree block, block, coded block, etc. that have been decoded, and inter-prediction that predicts from the image signal of the processed image. Switch (MODE_INTER). The mode that distinguishes between the intra prediction (MODE_INTRA) and the inter prediction (MODE_INTER) is defined as the prediction mode (PredMode). The prediction mode (PredMode) has an intra prediction (MODE_INTRA) or an inter prediction (MODE_INTER) as a value.

<Inter prediction>
In the inter-prediction that predicts from the image signal of the processed image, a plurality of processed images can be used as reference pictures. In order to manage a plurality of reference pictures, two types of reference lists, L0 (reference list 0) and L1 (reference list 1), are defined, and the reference pictures are specified by using the reference indexes of each. L0 prediction (Pred_L0) is available for P-slices. For B slices, L0 prediction (Pred_L0), L1 prediction (Pred_L1), and dual prediction (Pred_BI) can be used. The L0 prediction (Pred_L0) is an inter-prediction that refers to a reference picture managed by L0, and the L1 prediction (Pred_L1) is an inter-prediction that refers to a reference picture managed by L1. The bi-prediction (Pred_BI) is an inter-prediction in which both L0 prediction and L1 prediction are performed, and one reference picture managed in each of L0 and L1 is referred to. Information that identifies L0 prediction, L1 prediction, and bi-prediction is defined as an inter-prediction mode. In the subsequent processing, it is assumed that the constants and variables with the subscript LX attached to the output are processed for each L0 and L1.

<Predicted motion vector mode>
The predicted motion vector mode is a mode in which an index for specifying a predicted motion vector, a differential motion vector, an inter-prediction mode, and a reference index are transmitted to determine inter-prediction information of a block to be processed. The predicted motion vector is a predicted motion vector candidate derived from a processed block adjacent to the processed block or a block belonging to the processed image and located at the same position as or near (near) the processed block, and the predicted motion. Derived from the index to identify the vector.

<Merge mode>
In the merge mode, the processed block adjacent to the processed block or the block belonging to the processed image and located at the same position as or near (near) the processed block without transmitting the differential motion vector and the reference index. This mode derives the inter-prediction information of the block to be processed from the inter-prediction information of.

The processed block adjacent to the processing target block and the inter-prediction information of the processed block are defined as spatial merge candidates. Blocks that belong to the processed image and are located at the same position as or near (near) the block to be processed, and inter-prediction information derived from the inter-prediction information of that block are defined as time merge candidates. Each merge candidate is registered in the merge candidate list, and the merge index identifies the merge candidate used in the prediction of the block to be processed.

<Adjacent block>
FIG. 11 is a diagram illustrating a reference block referred to for deriving inter-prediction information in the predicted motion vector mode and the merge mode. A0, A1, A2, B0, B1, B2, and B3 are processed blocks adjacent to the processing target block. T0 is a block belonging to the processed image, and is a block located at the same position as or near (near) the processing target coding block of the processing target image.

A1 and A2 are blocks located on the left side of the processing target coding block and adjacent to the processing target coding block. B1 and B3 are blocks located above the processing target coding block and adjacent to the processing target coding block. A0, B0, and B2 are blocks located at the lower left, upper right, and upper left of the coded block to be processed, respectively.

The details of how to handle adjacent blocks in the predicted motion vector mode and the merge mode will be described later.

<Affine transformation motion compensation>
The affine transformation motion compensation divides a coded block into sub-blocks of a predetermined unit, sets a motion vector individually for each sub-block, and performs motion compensation. The motion vector of each sub-block is derived from the inter-prediction information of the processed block adjacent to the processed block or the block belonging to the processed image and located at the same position as or near (near) the processed block. Derived based on one or more control points. In the present embodiment, the size of the subblock is 4x4 pixels, but the size of the subblock is not limited to this, and the motion vector may be derived in pixel units.

FIG. 14 shows an example of affine transformation motion compensation when there are two control points. In this case, since the two control points have two parameters, a horizontal component and a vertical component, the affine transformation when there are two control points is called a four-parameter affine transformation. CP1 and CP2 in FIG. 14 are control points. FIG. 15 shows an example of affine transformation motion compensation when there are three control points. In this case, since the three control points have two parameters, a horizontal component and a vertical component, the affine transformation when there are three control points is called a six-parameter affine transformation. CP1, CP2, and CP3 in FIG. 15 are control points.

Affine transformation motion compensation is available in both the predicted motion vector mode and the merge mode. The mode in which the affine transformation motion compensation is applied in the predicted motion vector mode is defined as the subblock predicted motion vector mode, and the mode in which the affine transformation motion compensation is applied in the merge mode is defined as the subblock merge mode.

<Syntax of coded block>
The syntax for expressing the prediction mode of the coded block will be described with reference to FIGS. 12 (a), 12 (b), and 13. The pred_mode_flag in FIG. 12A is a flag indicating whether or not it is an inter-prediction. If pred_mode_flag is 0, it is an inter-prediction, and if pred_mode_flag is 1, it is an intra-prediction. In the case of intra-prediction, the information of intra-prediction intra_pred_mode is sent, and in the case of inter-prediction, merge_flag is sent. The merge_flag is a flag indicating whether to use the merge mode or the predicted motion vector mode. In the case of the predicted motion vector mode (merge_flag = 0), the flag inter_affine_flag indicating whether or not to apply the subblock predicted motion vector mode is sent. When applying the subblock prediction motion vector mode (inter_affine_flag = 1), send cu_affine_type_flag. cu_affine_type_flag is a flag for determining the number of control points in the subblock prediction motion vector mode.

On the other hand, in the case of merge mode (merge_flag = 1), the merge_subblock_flag shown in FIG. 12B is sent. merge_subblock_flag is a flag indicating whether or not to apply the subblock merge mode. In subblock merge mode (merge_subblock_flag = 1), send the merge index merge_subblock_idx. On the other hand, if it is not in subblock merge mode (merge_subblock_flag = 0), it sends a flag merge_triangle_flag indicating whether to apply the triangular merge mode. When applying the triangular merge mode (merge_triangle_flag = 1), send the merge_triangle_split_dir direction to divide the block and the merge triangular indexes merge_triangle_idx0 and merge_triangle_idx1 for each of the two divided partitions. On the other hand, if the triangular merge mode is not applied (merge_triangle_flag = 0), the merge index merge_idx is sent.

FIG. 13 shows the values of each syntax element and the corresponding prediction modes. merge_flag = 0 and inter_affine_flag = 0 correspond to the normal predicted motion vector mode (Inter Pred Mode). merge_flag = 0 and inter_affine_flag = 1 correspond to the subblock predicted motion vector mode (Inter Affine Mode). merge_flag = 1, merge_subblock_flag = 0, merge_trianlge_flag = 0 correspond to the normal merge mode (Merge Mode). merge_flag = 1, merge_subblock_flag = 0, merge_trianlge_flag = 1 correspond to Triangle Merge Mode. merge_flag = 1, merge_subblock_flag = 1 corresponds to the subblock merge mode (Affine Merge Mode).

<POC>
The POC (Picture Order Count) is a variable associated with the encoded picture, and a value that increases by 1 in the output order of the picture is set. Depending on the value of POC, it is possible to determine whether the pictures are the same, determine the context between the pictures in the output order, and derive the distance between the pictures. For example, if the POCs of the two pictures have the same value, it can be determined that they are the same picture. If the POCs of the two pictures have different values, it can be determined that the picture with the smaller POC value is the picture to be output first, and the difference between the POCs of the two pictures is the distance between the pictures in the time axis direction. Shown.

(First Embodiment)
The image coding device 100 and the image decoding device 200 according to the first embodiment of the present invention will be described.

FIG. 1 is a block diagram of the image coding apparatus 100 according to the first embodiment. The moving image coding device of the embodiment includes an image coding device 100, a block division unit 101, an inter prediction unit 102, an intra prediction unit 103, a decoded image memory 104, a prediction method determination unit 105, and a residual signal generation unit 106. It includes an orthogonal conversion / quantization unit 107, a bit string coding unit 108, an inverse quantization / inverse orthogonal conversion unit 109, a decoded image signal superimposition unit 110, and a coding information storage memory 111.

The block division unit 101 recursively divides the input image to generate a coded block. The block division unit 101 includes a 4-division unit that divides the block to be divided in the horizontal direction and the vertical direction, respectively, and a 2-3 division unit that divides the block to be divided into either the horizontal direction or the vertical direction. .. The generated image signal of the coded block to be processed is supplied to the inter prediction unit 102, the intra prediction unit 103, and the residual signal generation unit 106. Further, information indicating the determined recursive division structure is supplied to the bit string coding unit 108. The detailed operation of the block dividing unit 101 will be described later.

The inter-prediction unit 102 performs inter-prediction of the coded block to be processed. Multiple inter-prediction information candidates are derived from the inter-prediction information stored in the coded information storage memory and the decoded image signal stored in the decoded image memory 104, and suitable inter-prediction is derived from the plurality of candidates. A mode is selected, and the selected inter-prediction mode and the prediction image signal corresponding to the selected inter-prediction mode are supplied to the prediction method determination unit 105. The detailed configuration and operation of the inter-prediction unit 102 will be described later.

The intra prediction unit 103 performs intra prediction of the coded block to be processed. A predicted image signal is generated by intra prediction from the decoded image signal stored in the decoded image memory 104, a suitable intra prediction mode is selected from a plurality of intra prediction modes, and the selected intra prediction mode and the selected intra prediction mode are selected. The prediction image signal corresponding to the selected intra prediction mode is supplied to the prediction method determination unit 105. FIG. 10 shows an example of intra prediction. FIG. 10A shows the correspondence between the prediction direction of the intra prediction and the intra prediction mode number. For example, the intra prediction mode 50 generates an intra prediction image by copying pixels in the vertical direction. The intra prediction mode 1 is a DC mode, and is a mode in which all the pixel values of the processing target block are set as the average value of the reference pixels. The intra prediction mode 0 is a Planar mode, which is a mode for creating a two-dimensional intra prediction image from reference pixels in the vertical and horizontal directions. FIG. 10B is an example of generating an intra prediction image in the case of the intra prediction mode 40. For each pixel of the processing target block, the value of the reference pixel in the direction indicated by the intra prediction mode is copied. When the reference pixel in the intra prediction mode is not an integer position, the reference pixel value is determined by interpolation from the reference pixel values at the surrounding integer positions.

The decoded image memory 104 stores the decoded image generated by the decoded image signal superimposing unit 110. The decoded image stored in the decoded image memory is supplied to the inter prediction unit 102 and the intra prediction unit 103.

The prediction method determination unit 105 evaluates each prediction by using the coding information, the code amount of the residual signal, the amount of distortion between the predicted image signal and the image signal to be processed, and the like, thereby performing the optimum prediction. Determine the mode (inter-prediction or intra-prediction). In the case of the inter-prediction merge mode, the code information of the merge index and the information indicating whether or not the sub-block merge mode (sub-block merge flag) is supplied to the bit string coding unit 108, and the inter-prediction prediction motion vector mode is used. In the case, the bit string coding unit 108 provides coding information such as an inter-prediction mode, a prediction motion vector index, a reference index of L0 and L1, a difference motion vector, and information indicating whether or not the mode is a subblock mode (subblock prediction motion vector flag). Supply to. The determined coding information is supplied to the coding information storage memory 111.

The residual signal generation unit 106 generates a residual signal by subtracting the predicted image signal from the image signal to be processed, and supplies the residual signal to the orthogonal conversion / quantization unit 107.

The orthogonal conversion / quantization unit 107 performs orthogonal conversion and quantization on the residual signal according to the quantization parameter to generate an orthogonal conversion / quantization residual signal, and is inversely quantum with the bit string coding unit 108. It is supplied to the quantization / inverse orthogonal conversion unit 109.

The bit string coding unit 108 encodes the coding information according to the prediction method determined by the prediction method determination unit 105 for each coding block, in addition to the information in units of sequences, pictures, slices, and coding blocks. Specifically, in the case of prediction mode PredMode, division mode PartMode, and inter-prediction (PRED_INTER) for each coded block, a flag for determining whether or not it is in merge mode, a subblock merge flag, and in the case of merge mode, a merge index and merge. If it is not in the mode, the coding information such as the inter-prediction mode, the prediction motion vector index, the information about the difference motion vector, and the subblock prediction motion vector flag is encoded according to the specified syntax rules described later to generate the first encoded bit string. To do. Further, the bit string coding unit 108 entropy-codes the orthogonally converted and quantized residual signal according to a defined syntax rule to generate a second coded bit string. The first coded bit string and the second coded bit string are multiplexed according to the specified syntax rules, and a bit stream is output.

The inverse quantization / inverse orthogonal conversion unit 109 performs inverse quantization and inverse orthogonal conversion of the orthogonal conversion / quantization residual signal supplied from the orthogonal conversion / quantization unit 107 to calculate the residual signal and decode it. It is supplied to the image signal superimposing unit 110.

The decoded image signal superimposition unit 110 superimposes the predicted image signal according to the determination by the prediction method determination unit 105 and the residual signal that has been inversely quantized and inversely orthogonally converted by the inverse quantization / inverse orthogonal conversion unit 109 to obtain the decoded image. Is generated and stored in the decoded image memory 104. The decoded image may be stored in the decoded image memory 104 after being subjected to a filtering process for reducing distortion such as block distortion due to encoding.

The coding information storage memory 111 stores coding information such as a prediction mode (inter-prediction or intra-prediction) determined by the prediction method determination unit 105. In the case of inter-prediction, the coded information stored in the coded information storage memory 111 is the determined motion vector, reference list, and reference index, and in the case of inter-prediction merge mode, is it the merge index or subblock merge mode? Coding information of information indicating whether or not (sub-block merge flag), inter-prediction mode, predicted motion vector index in the case of inter-prediction predicted motion vector mode, reference index of L0, L1, differential motion vector, sub-block mode Information indicating whether or not (subblock prediction motion vector flag), in the case of intra prediction, the determined intra prediction mode and the like. The construction of the history candidate list managed by the coded information storage memory 111 will be described later.

FIG. 2 is a block showing a configuration of a moving image decoding device according to an embodiment of the present invention corresponding to the moving image coding device of FIG. The moving image decoding device of the embodiment includes a bit string decoding unit 201, a block division unit 202, an inter prediction unit 203, an intra prediction unit 204, a coded information storage memory 205, an inverse quantization / inverse orthogonal conversion unit 206, and a decoded image signal. A superimposing unit 207 and a decoded image memory 208 are provided.

Since the decoding process of the moving image decoding device of FIG. 2 corresponds to the decoding process provided inside the moving image coding device of FIG. 1, the coding information storage memory 205 of FIG. 2 is dequantized. The configurations of the inverse orthogonal conversion unit 206, the decoded image signal superimposing unit 207, and the decoded image memory 208 are the inverse quantization / inverse orthogonal conversion unit 109 of the moving image coding apparatus of FIG. 1, the decoded image signal superimposing unit 110, It has a function corresponding to each configuration of the coding information storage memory 111 and the decoded image memory 104.

The bit stream supplied to the bit string decoding unit 201 is separated according to a specified syntax rule. The separated first coded bit string is decoded to obtain sequence, picture, slice, coded block unit information, and coded block unit coded information. Specifically, in the case of prediction mode PredMode, split mode PartMode, and inter-prediction (PRED_INTER) that determine whether it is inter-prediction (PRED_INTER) or intra-prediction (PRED_INTRA) for each coded block, a flag that determines whether it is merge mode or not. , In the case of merge mode, merge index, sub-block merge flag, in the case of predicted motion vector mode, inter-predicted mode, predicted motion vector index, differential motion vector, sub-block predicted motion vector flag, etc. Decoding is performed according to the syntax rules of the above, and the coded information is supplied to the inter-prediction unit 203, the intra-prediction unit 204, and the coded information storage memory 205. The separated second coded bit string is decoded to calculate the orthogonally converted / quantized residual signal, and the orthogonally converted / quantized residual signal is supplied to the inversely quantized / inversely orthogonal converter 206.

The inter-prediction unit 203 describes the code of the already decoded image signal stored in the coding information storage memory 205 when the prediction mode PredMode of the coded block to be processed is the inter-prediction (PRED_INTER) and the prediction motion vector mode. Multiple prediction motion vector candidates are derived using the conversion information and registered in the prediction motion vector candidate list described later, and bit string decoding is performed from the plurality of prediction motion vector candidates registered in the prediction motion vector candidate list. The predicted motion vector corresponding to the predicted motion vector index decoded and supplied by the unit 201 is selected, the motion vector is calculated from the difference vector decoded by the bit string decoding unit 201 and the selected predicted motion vector, and other coding is performed. It is stored in the coded information storage memory 205 together with the information. The coding information of the coding block supplied and stored here is the flag predFlagL0 [xP] [yP], predFlagL1 [xP] indicating whether or not to use the prediction mode PredMode, the division mode PartMode, the L0 prediction, and the L1 prediction. Reference indexes of [yP], L0, L1 refIdxL0 [xP] [yP], refIdxL1 [xP] [yP], motion vectors mvL0 [xP] [yP], mvL1 [xP] [yP] of L0, L1 etc. .. Here, xP and yP are indexes indicating the positions of the upper left pixels of the coding block in the picture. When the prediction mode PredMode is inter-prediction (MODE_INTER) and the inter-prediction mode is L0 prediction (Pred_L0), the flag predFlagL0 indicating whether to use L0 prediction is 1, and the flag predFlagL1 indicating whether to use L1 prediction. Is 0. When the inter-prediction mode is L1 prediction (Pred_L1), the flag predFlagL0 indicating whether or not to use the L0 prediction is 0, and the flag predFlag L1 indicating whether or not to use the L1 prediction is 1. When the inter-prediction mode is bi-prediction (Pred_BI), the flag predFlagL0 indicating whether or not to use L0 prediction and the flag predFlagL1 indicating whether or not to use L1 prediction are both 1. Further, when the prediction mode PredMode of the coded block to be processed is inter-prediction (PRED_INTER) and the merge mode is set, merge candidates are derived. Using the coded information of the already decoded coded block stored in the coded information storage memory 205, a plurality of merge candidates are derived, registered in the merge candidate list described later, and registered in the merge candidate list. A flag indicating whether or not to use the L0 prediction and the L1 prediction of the selected merge candidate by selecting the merge candidate corresponding to the merge index decoded and supplied by the bit string decoding unit 201 from the plurality of merge candidates. PredFlagL0 [xP] [yP], predFlagL1 [xP] [yP], reference indexes refIdxL0 [xP] [yP], refIdxL1 [xP] [yP], L0, L1 motion vectors mvL0 [xP] [yP] ], MvL1 [xP] [yP] and other inter-prediction information is stored in the coded information storage memory 205. Here, xP and yP are indexes indicating the positions of the upper left pixels of the coding block in the picture. The detailed configuration and operation of the inter-prediction unit will be described later.

The intra prediction unit 204 performs intra prediction when the prediction mode PredMode of the coded block to be processed is intra prediction (PRED_INTRA). The coded information decoded by the bit string decoding unit 201 includes an intra prediction mode, and a predicted image signal is predicted by intra prediction from the decoded image signal stored in the decoded image memory 208 according to the intra prediction mode. Is generated, and the predicted image signal is supplied to the decoded image signal superimposing unit 207. Since the intra prediction unit 204 corresponds to the intra prediction unit 103 of the image coding device 100, the same processing as that of the intra prediction unit 103 is performed.

The inverse quantization / anti-orthogonal conversion unit 206 performs anti-orthogonal conversion and anti-quantization on the orthogonal conversion / quantization residual signal decoded by the bit string decoding unit 201, and is subjected to anti-orthogonal conversion / anti-quantization. Quantization signal is obtained.

The decoded image signal superimposition unit 207 is inversely quantized by the inter-prediction unit 203 or the intra-prediction predicted image signal by the intra-prediction unit 204, or the inverse-orthogonal conversion unit 206. The decoded image signal is decoded and stored in the decoded image memory 208 by superimposing the dequantized residual signal. When storing in the decoded image memory 208, the decoded image may be stored in the decoded image memory 208 after being subjected to a filtering process for reducing block distortion or the like due to coding.

Next, the operation of the block dividing unit 101 in the image coding apparatus 100 will be described. FIG. 3 is a flowchart showing an operation of dividing an image into tree blocks and further dividing each tree block. First, the input image is divided into tree blocks of a predetermined size (step S1001). Each tree block is scanned in a predetermined order, that is, in the order of raster scan (step S1002), and the inside of the tree block to be processed is divided (step S1003).

FIG. 7 is a flowchart showing a detailed operation of the division process in step S1003. First, it is determined whether or not to divide the block to be processed into four (step S1101).

When it is determined that the processing target block is divided into four, the processing target block is divided into four (step S1102). Each block obtained by dividing the block to be processed is scanned in the Z scan order, that is, in the order of upper left, upper right, lower left, and lower right (step S1103). FIG. 5 is an example of the Z scan order, and FIG. 601 of FIG. 6 is an example of dividing the processing target block into four. Numbers 0 to 3 of 601 in FIG. 6 indicate the order of processing. Then, the flowchart of FIG. 7 is recursively called for each block divided in step S1101.

If it is determined that the block to be processed is not divided into four, it is divided into 2-3 (step S1105).

FIG. 8 is a flowchart showing a detailed operation of the 2-3 division process of step S1105. First, it is determined whether or not the block to be processed is divided into 2-3, that is, whether or not to perform either 2-division or 3-division (step S1201).

If it is not determined that the block to be processed is divided into 2-3, that is, if it is determined not to be divided, the division is ended (step S1211), and the process returns to the upper layer block.

When it is determined that the block to be processed is divided into 2-3, it is determined whether or not to further divide the block to be processed into two (step S1202).

When it is determined that the processing target block is divided into two, it is determined whether or not to divide the processing target block in the vertical direction (step S1203), and based on the result, the processing target block is divided in the vertical direction (step S1204). Alternatively, the block to be processed is divided in the horizontal direction (step S1205). As a result of step S1204, the processing target block is divided into two vertical divisions as shown in FIG. 602, and as a result of step S1205, the processing target block is divided into two horizontal divisions as shown in FIG. 604.

In step S1202, if it is not determined that the block to be processed is divided into two, that is, if it is determined to be divided into three, it is determined whether or not to divide the block to be processed in the vertical direction (step S1206), and the result is Based on this, the processing target block is divided in the vertical direction (step S1207), or the processing target block is divided in the horizontal direction (step S1208). As a result of step S1207, the processing target block is divided into three vertical divisions as shown in FIG. 603, and as a result of step S1208, the processing target block is divided into three horizontal divisions as shown in FIG. 605.

After executing any of steps S1204 to S1205, each block obtained by dividing the processing target block is scanned in the order of left to right and top to bottom (step S1209). Numbers 0 to 3 of 602 to 605 in FIG. 6 indicate the order of processing. The flowchart of FIG. 8 is recursively called for each divided block.

In the recursive block division described here, the necessity of division may be limited depending on the number of divisions, the size of the block to be processed, and the like. The information that limits the necessity of division may be realized in a configuration that does not transmit information by making an agreement in advance between the coding device and the decoding device, or the coding device limits the necessity of division. It may be realized by the configuration which transmits to the decoding apparatus by determining the information to be performed and recording it in the coded bit string.

Here, when a certain block is divided, the block before the division is called a parent block, and each block after the division is called a child block.

Next, the operation of the block dividing unit 202 in the image decoding device 200 will be described. The block division unit 202 divides the tree block by the same processing procedure as the block division unit 101 of the image coding apparatus 100. However, the block division unit 101 of the image coding apparatus 100 applies optimization methods such as estimation of the optimum shape by image recognition and optimization of the distortion rate to determine the optimum shape of the block division, whereas the image decoding apparatus The block division unit 202 in 200 is different in that the block division shape is determined by decoding the block division information recorded in the coded bit string.

The syntax (syntax rule of the coded bit string) regarding the block division of the first embodiment is shown in FIG. coding_quadtree () represents the syntax for the block 4-split processing, and multi_type_tree () represents the syntax for the block 2-split or 3-split processing. qt_split is a flag indicating whether or not to divide the block into four. If the block is divided into four, qt_split = 1, and if not divided into four, qt_split = 0. When dividing into four (qt_split = 1), each block divided into four is recursively divided into four (coding_quadtree (0), coding_quadtree (1), coding_quadtree (2), coding_quadtree (3)). When not dividing into four (qt_split = 0), the subsequent division is determined according to multi_type_tree (). mtt_split is a flag indicating whether or not to further divide. When further dividing (mtt_split = 1), refer to mtt_split_vertical, which is a flag indicating whether to divide vertically or horizontally, and mtt_split_binary, which is a flag that determines whether to divide into two or three. mtt_split_vertical = 1 indicates splitting in the vertical direction, and mtt_split_vertical = 0 indicates splitting in the horizontal direction. mtt_split_binary = 1 indicates that it is divided into two, and mtt_split_binary = 0 indicates that it is divided into three. Hierarchical block division is performed by recursively calling multi_type_tree until mtt_split = 0.

<Inter prediction>
The inter-prediction method according to the embodiment is carried out by the inter-prediction unit 102 of the moving image coding device of FIG. 1 and the inter-prediction unit 203 of the moving image decoding device of FIG.

The inter-prediction method according to the embodiment will be described with reference to the drawings. The inter-prediction method is performed in either coding or decoding processing in units of coded blocks.

<Explanation of Inter Prediction Unit 102 on the Encoding Side>
FIG. 16 is a diagram showing a detailed configuration of the inter-prediction unit 102 of the moving image coding device of FIG. The normal predicted motion vector mode derivation unit 301 derives a plurality of normal predicted motion vector candidates, selects a predicted motion vector, and calculates a difference vector from the detected motion vector. The detected inter-prediction mode, reference index, motion vector, and calculated difference vector become the inter-prediction information of the normal prediction motion vector mode. This inter-prediction information is supplied to the inter-prediction mode determination unit 305. The detailed configuration and processing of the normal prediction motion vector mode derivation unit 301 will be described later.

The normal merge mode derivation unit 302 derives a plurality of normal merge candidates, selects the normal merge candidates, and obtains inter-prediction information in the normal merge mode. This inter-prediction information is supplied to the inter-prediction mode determination unit 305. The detailed configuration and processing of the normal merge mode derivation unit 302 will be described later.

The sub-block prediction motion vector mode derivation unit 303 derives a plurality of sub-block prediction motion vector candidates, selects a sub-block prediction motion vector, and calculates a difference vector from the detected motion vector. The detected inter-prediction mode, reference index, motion vector, and calculated difference vector become the inter-prediction information of the normal prediction motion vector mode. This inter-prediction information is supplied to the inter-prediction mode determination unit 305. The detailed configuration and processing of the subblock prediction motion vector mode derivation unit 303 will be described later.

The sub-block merge mode derivation unit 304 derives a plurality of sub-block merge candidates, selects the sub-block merge candidates, and obtains the inter-prediction information of the sub-block merge mode. This inter-prediction information is supplied to the inter-prediction mode determination unit 305. The detailed configuration and processing of the subblock merge mode derivation unit 304 will be described later.

The inter prediction mode determination unit 305 is based on the inter prediction information supplied from the normal prediction motion vector mode derivation unit 301, the normal merge mode derivation unit 302, the subblock prediction motion vector mode derivation unit 303, and the subblock merge mode derivation unit 304. , Determine the inter-prediction mode. Inter-prediction mode determination unit 305 supplies inter-prediction information according to the determination result to motion compensation prediction unit 306.

The motion compensation prediction unit 306 performs inter-prediction for the reference image signal stored in the decoded image memory 104 based on the determined inter-prediction information. The detailed configuration and processing will be described later.

<Explanation of inter-prediction unit 203 on the decoding side>
FIG. 22 is a diagram showing a detailed configuration of the inter-prediction unit 203 of the moving image decoding device of FIG.

The normal predicted motion vector mode derivation unit 401 derives a plurality of normal predicted motion vector candidates, selects a predicted motion vector, and calculates a difference vector from the detected motion vector. The detected inter-prediction mode, reference index, motion vector, and difference vector are the inter-prediction information of the normal prediction motion vector mode. This inter-prediction information is supplied to the motion compensation prediction unit 406 via the switch 408. The detailed configuration and processing of the normal predicted motion vector mode derivation unit 401 will be described later.

The normal merge mode derivation unit 402 derives a plurality of normal merge candidates, selects the normal merge candidates, and obtains inter-prediction information in the normal merge mode. This inter-prediction information is supplied to the motion compensation prediction unit 406 via the switch 408. The detailed configuration and processing of the normal merge mode derivation unit 402 will be described later.

The sub-block prediction motion vector mode derivation unit 403 derives a plurality of sub-block prediction motion vector candidates, selects a sub-block prediction motion vector, and calculates a difference vector from the detected motion vector. The detected inter-prediction mode, reference index, motion vector, and calculated difference vector become the inter-prediction information of the normal prediction motion vector mode. This inter-prediction information is supplied to the motion compensation prediction unit 406 via the switch 408. The detailed configuration and processing of the subblock prediction motion vector mode derivation unit 403 will be described later.

The sub-block merge mode derivation unit 404 derives a plurality of sub-block merge candidates, selects the sub-block merge candidates, and obtains inter-prediction information of the sub-block merge mode. This inter-prediction information is supplied to the motion compensation prediction unit 406 via the switch 408. The detailed configuration and processing of the subblock merge mode derivation unit 404 will be described later.

The motion compensation prediction unit 406 performs inter-prediction for the reference image signal stored in the decoded image memory 208 based on the determined inter-prediction information. The detailed configuration and processing are the same as on the coding side.

<Normal prediction motion vector mode derivation unit (normal AMVP)>
The normal prediction motion vector mode derivation unit 301 of FIG. 17 includes a spatial prediction motion vector candidate derivation unit 321, a time prediction motion vector candidate derivation unit 322, a history prediction motion vector candidate derivation unit 323, a prediction motion vector candidate replenishment unit 325, and a normal motion. It includes a vector detection unit 326, a predicted motion vector candidate selection unit 327, and a motion vector subtraction unit 328.

The normal prediction motion vector mode derivation unit 401 of FIG. 23 includes a space prediction motion vector candidate derivation unit 421, a time prediction motion vector candidate derivation unit 422, a history prediction motion vector candidate derivation unit 423, a prediction motion vector candidate replenishment unit 425, and a prediction motion. It includes a vector candidate selection unit 426 and a motion vector addition unit 427.

The processing procedure of the normal prediction motion vector mode derivation unit 301 on the coding side and the normal prediction motion vector mode derivation unit 401 on the decoding side will be described with reference to the flowcharts of FIGS. 19 and 25, respectively. FIG. 19 is a flowchart showing a normal motion vector mode derivation processing procedure by the normal motion vector mode derivation unit 301 on the coding side, and FIG. 25 is a normal motion vector mode derivation process by the normal motion vector mode derivation unit 401 on the decoding side. It is a flowchart which shows the procedure.

<Normal prediction motion vector mode derivation unit (normal AMVP): Explanation on the coding side>
The procedure for deriving the normal predicted motion vector mode on the coding side will be described with reference to FIG. In the description of the processing procedure of FIG. 19, the term motion vector in the specification and the term normal motion vector of FIG. 19 correspond to each other. First, the normal motion vector detection unit 326 detects the normal motion vector for each inter-prediction mode and reference index (step S100 in FIG. 19).

Subsequently, the spatial prediction motion vector candidate derivation unit 321, the time prediction motion vector candidate derivation unit 322, the history prediction motion vector candidate derivation unit 323, the prediction motion vector candidate supplementation unit 325, the prediction motion vector candidate selection unit 327, and the motion vector subtraction unit. At 328, the difference motion vector of the motion vector used in the inter-prediction of the normal prediction motion vector mode is calculated for each L0 and L1, respectively (steps S101 to S106 in FIG. 19). Specifically, when the prediction mode PredMode of the block to be processed is inter-prediction (MODE_INTER) and the inter-prediction mode is L0 prediction (Pred_L0), the prediction motion vector candidate list mvpListL0 of L0 is calculated and the prediction motion vector mvpL0 is selected. Then, the difference motion vector mvdL0 of the motion vector mvL0 of L0 is calculated. When the inter-prediction mode of the block to be processed is L1 prediction (Pred_L1), the prediction motion vector candidate list mvpListL1 of L1 is calculated, the prediction motion vector mvpL1 is selected, and the difference motion vector mvdL1 of the motion vector mvL1 of L1 is calculated. .. When the inter-prediction mode of the block to be processed is bi-prediction (Pred_BI), both L0 prediction and L1 prediction are performed, the prediction motion vector candidate list mvpList L0 of L0 is calculated, the prediction motion vector mvpL0 of L0 is selected, and L0. The motion vector mvL0 of the motion vector mvL0 is calculated, the predicted motion vector candidate list mvpListL1 of L1 is calculated, the predicted motion vector mvpL1 of L1 is calculated, and the differential motion vector mvdL1 of the motion vector mvL1 of L1 is calculated. To do.

The difference motion vector calculation process is performed for each of L0 and L1, but both L0 and L1 are common processes. Therefore, in the following description, L0 and L1 are represented as a common LX. X is 0 in the process of calculating the differential motion vector of L0, and X is 1 in the process of calculating the differential motion vector of L1. Further, when the information of the other list is referred to instead of the LX during the process of calculating the differential motion vector of the LX, the other list is represented as LY.

When the motion vector mvLX of LX is used (step S102: YES in FIG. 19), the candidates of the predicted motion vector of LX are calculated to construct the predicted motion vector candidate list mvpListLX of LX (step S103 of FIG. 19). Multiple predicted motions in the space predicted motion vector candidate derived section 321 in the normal predicted motion vector mode derived section 301, the time predicted motion vector candidate derived section 322, the historical predicted motion vector candidate derived section 323, and the predicted motion vector candidate supplement section 325. Derivation of vector candidates and construction of predicted motion vector candidate list mvpListLX. The detailed processing procedure of step S103 of FIG. 19 will be described later using the flowchart of FIG.

Subsequently, the predicted motion vector candidate selection unit 327 selects the LX predicted motion vector mvpLX from the LX predicted motion vector candidate list mvpListLX (step S104 in FIG. 19). Each difference motion vector, which is the difference between the motion vector mvLX and the candidate mvpListLX [i] of each predicted motion vector stored in the predicted motion vector candidate list mvpListLX, is calculated. The amount of code when these differential motion vectors are encoded is calculated for each element of the predicted motion vector candidate list mvpListLX. Then, among the elements registered in the predicted motion vector candidate list mvpListLX, the predicted motion vector candidate mvpListLX [i] having the minimum sign amount for each candidate of the predicted motion vector is selected as the predicted motion vector mvpLX, and the candidate mvpListLX [i] is selected. Get index i. When there are multiple candidates for the predicted motion vector that is the smallest generated code amount in the predicted motion vector candidate list mvpListLX, the index i in the predicted motion vector candidate list mvpListLX is represented by a small number. Select the candidate mvpListLX [i] as the optimal predicted motion vector mvpLX and get its index i.

Subsequently, the motion vector subtraction unit 328 subtracts the selected LX predicted motion vector mvpLX from the LX motion vector mvLX.
mvdLX = mvLX --mvpLX
The difference motion vector mvdLX of LX is calculated as (step S105 in FIG. 19).

<Normal prediction motion vector mode derivation unit (normal AMVP): Explanation on the decoding side>
Next, the normal predicted motion vector mode processing procedure on the decoding side will be described with reference to FIG. On the decoding side, the spatial prediction motion vector candidate derivation unit 421, the time prediction motion vector candidate derivation unit 422, the history prediction motion vector candidate derivation unit 423, and the prediction motion vector candidate replenishment unit 425 are used for inter-prediction in the normal prediction motion vector mode. The motion vector is calculated for each L0 and L1 (steps S201 to S206 in FIG. 25). Specifically, when the prediction mode PredMode of the processing target block is inter-prediction (MODE_INTER) and the inter-prediction mode of the processing target block is L0 prediction (Pred_L0), the prediction motion vector candidate list mvpListL0 of L0 is calculated and the prediction motion is predicted. Select the vector mvpL0 and calculate the motion vector mvL0 of L0. When the inter-prediction mode of the block to be processed is L1 prediction (Pred_L1), the prediction motion vector candidate list mvpListL1 of L1 is calculated, the prediction motion vector mvpL1 is selected, and the motion vector mvL1 of L1 is calculated. When the inter-prediction mode of the block to be processed is bi-prediction (Pred_BI), both L0 prediction and L1 prediction are performed, the prediction motion vector candidate list mvpList L0 of L0 is calculated, the prediction motion vector mvpL0 of L0 is selected, and L0. The motion vector mvL0 of L1 is calculated, the predicted motion vector candidate list mvpList L1 of L1 is calculated, the predicted motion vector mvpL1 of L1 is calculated, and the motion vector mvL1 of L1 is calculated respectively.

Similar to the coding side, the decoding side also performs motion vector calculation processing for each of L0 and L1, but both L0 and L1 are common processing. Therefore, in the following description, L0 and L1 are represented as a common LX. LX represents an inter-prediction mode used for inter-prediction of the coded block to be processed. X is 0 in the process of calculating the motion vector of L0, and X is 1 in the process of calculating the motion vector of L1. Further, when the information of the other reference list is referred to instead of the same reference list as the LX to be calculated during the process of calculating the motion vector of the LX, the other reference list is represented as LY.

When the motion vector mvLX of LX is used (step S202: YES in FIG. 25), the candidates of the predicted motion vector of LX are calculated to construct the predicted motion vector candidate list mvpListLX of LX (step S203 of FIG. 25). Multiple predicted motions in the space predicted motion vector candidate derived section 421, the time predicted motion vector candidate derived section 422, the historical predicted motion vector candidate derived section 423, and the predicted motion vector candidate supplement section 425 in the normal predicted motion vector mode derivation section 401. Calculate vector candidates and build a predicted motion vector candidate list mvpListLX. The detailed processing procedure of step S203 of FIG. 25 will be described later using the flowchart of FIG.

Subsequently, the predicted motion vector candidate selection unit 426 selects the predicted motion vector candidate mvpListLX [mvpIdxLX] corresponding to the index mvpIdxLX of the predicted motion vector decoded and supplied by the bit string decoding unit 201 from the predicted motion vector candidate list mvpListLX. It is taken out as the predicted motion vector mvpLX (step S204 in FIG. 25).

Subsequently, the motion vector addition unit 427 adds the LX differential motion vector mvdLX and the LX predicted motion vector mvpLX, which are decoded and supplied by the bit string decoding unit 201.
mvLX = mvpLX + mvdLX
The motion vector mvLX of LX is calculated as (step S205 in FIG. 25).

<Normal prediction motion vector mode derivation unit (normal AMVP): Motion vector prediction method>
FIG. 20 shows a normal predicted motion vector having a function common to the normal predicted motion vector mode derivation unit 301 of the moving image coding device and the normal predicted motion vector mode derivation unit 401 of the moving image decoding device according to the embodiment of the present invention. It is a flowchart which shows the processing procedure of a mode derivation process.

The normal prediction motion vector mode derivation unit 301 and the normal prediction motion vector mode derivation unit 401 include a prediction motion vector candidate list mvpListLX. The predicted motion vector candidate list mvpListLX has a list structure, and is provided with a predicted motion vector index indicating the location inside the predicted motion vector candidate list and a storage area for storing the predicted motion vector candidates corresponding to the index as elements. The number of the predicted motion vector index starts from 0, and the predicted motion vector candidate is stored in the storage area of the predicted motion vector candidate list mvpListLX. In the present embodiment, the predicted motion vector candidate list mvpListLX can register at least two predicted motion vector candidates (inter-prediction information). Further, 0 is set in the variable numCurrMvpCand indicating the number of predicted motion vector candidates registered in the predicted motion vector candidate list mvpListLX.

Spatial prediction motion vector candidate derivation units 321 and 421 derive prediction motion vector candidates from blocks adjacent to the left side. In this process, the flag availableFlagLXA indicating whether or not the predicted motion vector candidate of the block (A0 or A1) adjacent to the left side is available, the motion vector mvLXA, and the reference index refIdxA are derived, and mvLXA is derived from the predicted motion vector candidate list mvpListLX. (Step S301 in FIG. 20). When L0, X is 0, and when L1, X is 1 (the same applies hereinafter). Subsequently, the spatial prediction motion vector candidate derivation units 321 and 421 derive the prediction motion vector candidates from the blocks (B0, B1 or B2) adjacent to the upper side. In this process, the flag availableFlagLXB indicating whether the predicted motion vector candidates of the adjacent blocks on the upper side are available, the motion vector mvLXB, and the reference index refIdxB are derived, and if mvLXA and mvLXB are not equal, mvLXB is predicted motion vector. Add to the candidate list mvpListLX (step S302 in FIG. 20). The processing of steps S301 and S302 in FIG. 20 is common except that the position and number of adjacent blocks to be referred to are different, and the flag availableFlagLXN indicating whether or not the predicted motion vector candidate of the coded block can be used, and the motion vector mvLXN. , The reference index refIdxN (N is A or B, and so on) is derived.

Subsequently, the time prediction motion vector candidate derivation units 322 and 422 derive the prediction motion vector candidates from the coding block in the picture whose time is different from the current processing target picture. In this process, the flags availableFlagLXCol indicating whether the predicted motion vector candidates of the coded blocks in the pictures at different times are available, the motion vector mvLXCol, the reference index refIdxCol, and the reference list listCol are derived, and the mvLXCol is used as the predicted motion vector candidate. Add to list mvpListLX (step S303 in FIG. 20). The derivation processing procedure of step S303 will be described in detail later.

It should be noted that the processing of the time prediction motion vector candidate derivation units 322 and 422 can be omitted in units of sequence (SPS), picture (PPS), or slice.

Subsequently, the history prediction motion vector candidate derivation units 323 and 423 add the history prediction motion vector candidates registered in the history prediction motion vector candidate list HmvpCandList to the prediction motion vector candidate list mvpListLX. (Step S304 in FIG. 20). The registration processing procedure in step S304 will be described in detail later with reference to the flowchart of FIG.

Subsequently, the predicted motion vector candidate supplementing units 325 and 425 add motion vectors having predetermined values such as until the predicted motion vector candidate list mvpListLX is satisfied (0,0) (S305 in FIG. 20).

<Normal merge mode derivation section (normal merge)>
The normal merge mode derivation unit 302 of FIG. 18 includes a spatial merge candidate derivation unit 341, a time merge candidate derivation unit 342, an average merge candidate derivation unit 344, a history merge candidate derivation unit 345, a merge candidate replenishment unit 346, and a merge candidate selection unit 347. including.

The normal merge mode derivation unit 402 of FIG. 24 includes a spatial merge candidate derivation unit 441, a time merge candidate derivation unit 442, an average merge candidate derivation unit 444, a history merge candidate derivation unit 445, a merge candidate replenishment unit 446, and a merge candidate selection unit 447. including.

FIG. 21 shows a procedure of the normal merge mode derivation process having a function common to the normal merge mode derivation unit 302 of the moving image coding device and the normal merge mode derivation unit 402 of the moving image decoding device according to the embodiment of the present invention. It is a flowchart to explain.

The various processes will be described below in order. In the following description, unless otherwise specified, the case where the slice type slice_type is B slice is described, but it can also be applied to the case of P slice. However, when the slice type slice_type is P slice, there is only L0 prediction (Pred_L0) as the inter-prediction mode, and there is no L1 prediction (Pred_L1) and double prediction (Pred_BI), so that the processing related to L1 can be omitted.

The normal merge mode derivation unit 302 and the normal merge mode derivation unit 402 include a merge candidate list mergeCandList. Merge candidate list mergeCandList has a list structure, and has a merge index indicating the location inside the merge candidate list and a storage area for storing merge candidates corresponding to the index as elements. The number of the merge index starts from 0, and the merge candidates are stored in the storage area of the merge candidate list mergeCandList. In the subsequent processing, the merge candidate of the merge index i registered in the merge candidate list mergeCandList is represented by mergeCandList [i]. In the present embodiment, the merge candidate list mergeCandList can register at least 6 merge candidates (inter-prediction information). Further, 0 is set in the variable numCurrMergeCand indicating the number of merge candidates registered in the merge candidate list mergeCandList.

The space merge candidate derivation unit 341 and the space merge candidate derivation unit 441 process from the coding information stored in the coding information storage memory 111 of the moving image coding device or the coding information storage memory 205 of the moving image decoding device. Spatial merge candidates A and B from blocks adjacent to the left and upper sides of the target block are derived, and the derived spatial merge candidates are registered in the merge candidate list mergeCandList (step S401 in FIG. 21). Here, N indicating either the spatial merge candidates A and B or the time merge candidate Col is defined. Flags availableFlagN indicating whether or not the inter-prediction information of block N can be used as spatial merge candidate N, reference index refIdxL0N of L0 of spatial merge candidate N and reference index refIdxL1N of L1 and L0 indicating whether L0 prediction is performed. The prediction flags predFlagL0N and the motion vector mvL0N of the L1 prediction flags predFlagL1N and L0 indicating whether or not the L1 prediction is performed, and the motion vector mvL1N of L1 are derived. However, in the present embodiment, since the merge candidate is derived without referring to other coded blocks included in the block including the coded block to be processed, it is included in the block including the coded block to be processed. Spatial merge candidates are not derived.

Subsequently, the time merge candidate derivation unit 342 and the time merge candidate derivation unit 442 derive the time merge candidates from the pictures at different times and register the derived time merge candidates in the merge candidate list mergeCandList (FIG. 21). Step S402). Flags availableFlagCol indicating whether time merge candidates are available, L0 prediction flags predFlagL0Col indicating whether L0 prediction of time merge candidates is performed, and L1 prediction flags predFlagL1Col indicating whether L1 prediction is performed, and L0. The motion vector mvL0Col and the motion vector mvL1Col of L1 are derived. The detailed processing procedure of step S402 will be described in detail later with reference to FIG. 56.

It should be noted that the processing of the time merge candidate derivation unit 342 and the time merge candidate derivation unit 442 can be omitted in units of sequence (SPS), picture (PPS), or slice.

Subsequently, the history merge candidate derivation unit 345 and the history merge candidate derivation unit 445 add the history prediction motion vector candidates registered in the history prediction motion vector candidate list HmvpCandList to the merge candidate list mergeCandList (step S403 in FIG. 21). .. The detailed processing procedure of step S403 will be described in detail later with reference to the flowchart of FIG.

Subsequently, the average merge candidate derivation unit 344 and the average merge candidate derivation unit 444 derive the average merge candidate from the merge candidate list mergeCandList and register the derived average merge candidate in the merge candidate list mergeCandList (step of FIG. 21). S404). The detailed processing procedure of step S404 will be described in detail later with reference to the flowchart of FIG.

Subsequently, in the merge candidate replenishment unit 346 and the merge candidate replenishment unit 446, if the number of merge candidates numCurrMergeCand registered in the merge candidate list mergeCandList is smaller than the maximum number of merge candidates MaxNumMergeCand, they are registered in the merge candidate list mergeCandList. The number of merge candidates numCurrMergeCand is the maximum number of merge candidates MaxNumMergeCand is used as the upper limit to derive additional merge candidates and register them in the merge candidate list mergeCandList (step S405 in FIG. 21). With the maximum number of merge candidates MaxNumMergeCand as the upper limit, in the P slice, merge candidates whose motion vector has a value of (0,0) at different reference indexes and whose prediction mode is L0 prediction (Pred_L0) are added. In the B slice, a merge candidate whose motion vector has a value of (0,0) at a different reference index and whose prediction mode is bi-prediction (Pred_BI) is added.

Subsequently, the merge candidate selection unit 347 and the merge candidate selection unit 447 select the merge candidate from the merge candidates registered in the merge candidate list mergeCandList. The merge candidate selection unit 347 on the coding side selects the merge candidate by calculating the code amount and the strain amount, and sends the merge index indicating the selected merge candidate and the inter-prediction information of the merge candidate to the motion compensation prediction unit 306. Supply. On the other hand, the merge candidate selection unit 447 on the decoding side selects the merge candidate based on the decoded merge index, and supplies the selected merge candidate to the motion compensation prediction unit 406.

In the normal merge mode derivation unit 302 and the normal merge mode derivation unit 402, when the size (product of width and height) of a certain coded block is less than 32, merge candidates are derived in the parent block of the coded block. Then, in all the child blocks, the merge candidates derived in the parent block are used. However, this is limited to cases where the size of the parent block is 32 or more and fits within the screen.

<Derivation of subblock prediction motion vector mode>
Subblock prediction motion vector mode derivation will be described.

FIG. 26 is a block diagram of the sub-block prediction motion vector mode derivation unit 303 in the coding apparatus of the present embodiment.

First, the affine inheritance prediction motion vector candidate derivation unit 361 derives the affine inheritance prediction motion vector candidate. Details of affine inheritance prediction motion vector candidate derivation will be described later.

Subsequently, the affine construction prediction motion vector candidate derivation unit 362 derives the affine construction prediction motion vector candidate. The details of deriving the affine construction prediction motion vector candidate will be described later.

Subsequently, the affine identical prediction motion vector candidate deriving unit 363 derives the affine identical prediction motion vector candidate. Details of deriving affine identical prediction motion vector candidates will be described later.

The subblock motion vector detection unit 366 detects a subblock motion vector suitable for the subblock prediction motion vector mode, and supplies the detected vector to the subblock prediction motion vector candidate selection unit 367 and the difference calculation unit 368.

The subblock prediction motion vector candidate selection unit 367 is a subblock prediction motion vector candidate derived by the affine inheritance prediction motion vector candidate derivation unit 361, the affine construction prediction motion vector candidate derivation unit 362, and the affine same prediction motion vector candidate derivation unit 363. Subblock prediction motion vector candidates are selected based on the motion vector supplied from the subblock motion vector detection unit 366, and information on the selected subblock prediction motion vector candidates is provided in the inter-prediction mode determination unit 305, It is supplied to the difference calculation unit 368.

The difference calculation unit 368 interprets the difference prediction motion vector obtained by subtracting the subblock prediction motion vector selected by the subblock prediction motion vector candidate selection unit 367 from the motion vector vector supplied from the subblock motion vector detection unit 366. It is supplied to the prediction mode determination unit 305.

FIG. 27 is a block diagram of the sub-block prediction motion vector mode derivation unit 403 in the decoding device of the present embodiment.

First, the affine inheritance prediction motion vector candidate derivation unit 461 derives the affine inheritance prediction motion vector candidate. The processing of the affine inheritance prediction motion vector candidate derivation unit 461 is the same as the processing of the affine inheritance prediction motion vector candidate derivation unit 361 in the coding apparatus of this embodiment.

Subsequently, the affine construction prediction motion vector candidate derivation unit 462 derives the affine construction prediction motion vector candidate. The processing of the affine construction prediction motion vector candidate derivation unit 462 is the same as the processing of the affine construction prediction motion vector candidate derivation unit 362 in the coding apparatus of the present embodiment.

Subsequently, the affine identical prediction motion vector candidate derivation unit 463 derives the affine identical prediction motion vector candidate. The processing of the affine identical prediction motion vector candidate derivation unit 463 is the same as the processing of the affine identical prediction motion vector candidate derivation unit 363 in the coding apparatus of this embodiment.

The sub-block prediction motion vector candidate selection unit 466 is a sub-block prediction motion vector candidate derived unit 461, an affine construction prediction motion vector candidate derivation unit 462, and an affine same prediction motion vector candidate derivation unit 463. Subblock prediction motion vector candidates are selected based on the prediction motion vector index transmitted and decoded from the encoding device, and information about the selected subblock prediction motion vector candidates is added to the motion compensation prediction unit 406. It is supplied to the calculation unit 467.

The addition calculation unit 467 adds a motion vector generated by adding a differential motion vector transmitted and decoded from the encoding device to the subblock prediction motion vector selected by the subblock prediction motion vector candidate selection unit 466, and motion compensation prediction. Supply to unit 406.

<Derivation of affine inheritance prediction motion vector candidates>
The affine inheritance prediction motion vector candidate derivation unit 361 will be described. The affine inheritance prediction motion vector candidate derivation unit 461 is the same as the affine inheritance prediction motion vector candidate derivation unit 361.

Affine inheritance The motion vector candidate inherits the motion vector information of the control point.

FIG. 30 is a diagram for explaining affine inheritance prediction motion vector candidate derivation.

The affine inheritance prediction motion vector candidate is obtained by searching the motion vector of the control point possessed by the spatially adjacent coded / decoded blocks.

Specifically, a maximum of one affine mode is searched for from the blocks (A0, A1) adjacent to the left side of the block to be processed and the blocks (B0, B1, B2) adjacent to the upper side of the block to be processed. , Affine inheritance prediction motion vector.

FIG. 34 is a flowchart for deriving affine inheritance prediction motion vector candidates.

First, the blocks (A0, A1) adjacent to the left side of the block to be processed are set as the left group (step S3101), and it is determined whether or not the block including A0 is a block using affine transformation motion compensation (affine mode). (Step S3102). When A0 is in the affine mode (step S3102: YES), the affine mode used by A0 is acquired (step S3103), and the process proceeds to the processing of the block adjacent to the upper side. When A0 is not in the affine mode (step S3102: NO), the target for deriving the affine inheritance prediction motion vector candidate is set to A0-> A1, and acquisition of the affine mode is attempted from the block including A1.

Subsequently, the blocks (B0, B1, B2) adjacent to the upper side of the block to be processed are set as the upper group (step S3104), and it is determined whether or not the block including B0 is in the affine mode (step S3105). When B0 is in the affine mode (step S3105: YES), the affine mode used by B0 is acquired (step S3106), and the process ends. When B0 is not in the affine mode (step S3105: NO), the target for deriving the affine inheritance prediction motion vector candidate is set to B0-> B1, and acquisition of the affine mode is attempted from the block including B1. Further, when B1 is not in the affine mode (step S3105: NO), the target for deriving the affine inheritance prediction motion vector candidate is set to B1-> B2, and the acquisition of the affine mode is attempted from the block including B2.

In this way, the groups are divided into the left block and the upper block, and the affine mode is searched for the left block in the order from the lower left to the upper left block, and for the left block, the affine mode is searched for in the order from the upper right to the upper left block. Therefore, it is possible to acquire two affine modes that are as different as possible, and it is possible to derive an affine prediction motion vector candidate in which any one of the affine prediction motion vectors has a smaller difference motion vector.

<Derivation of affine construction prediction motion vector candidates>
The affine construction prediction motion vector candidate derivation unit 362 will be described. The affine construction prediction motion vector candidate derivation unit 462 is the same as the affine construction prediction motion vector candidate derivation unit 362.

The affine construction prediction motion vector candidate constructs motion vector information of control points from motion information of spatially adjacent blocks.

FIG. 31 is a diagram for explaining the derivation of affine construction prediction motion vector candidates.

The affine construction predicted motion vector candidate can be obtained by constructing a new affine mode by combining motion vectors of spatially adjacent encoded / decoded blocks.

Specifically, the motion vector of the upper left control point CP0 is derived from the blocks (B2, B3, A2) adjacent to the upper left side of the block to be processed, and the blocks (B1, B0) adjacent to the upper right side of the block to be processed are derived. ), The motion vector of the upper right control point CP1 is derived, and the motion vector of the lower left control point CP2 is derived from the blocks (A1, A0) adjacent to the lower left side of the block to be processed.

FIG. 35 is a flowchart for deriving affine construction prediction motion vector candidates.

First, the upper left control point CP0, the upper right control point CP1, and the lower left control point CP2 are derived (step S3201). The upper left control point CP0 is calculated by searching for a reference block having the same reference image as the block to be processed in the order of priority of the B2, B3, and A2 reference blocks. The upper right control point CP1 is calculated by searching for a reference block having the same reference image as the block to be processed in the order of priority of the B1 and B0 reference blocks. The lower left control point CP2 is calculated by searching for a reference block having the same reference image as the block to be processed in the order of priority of the A1 and A0 reference blocks.

When the three control points mode is selected as the affine construction prediction motion vector (step S3202: YES), it is determined whether or not all three control points (CP0, CP1, CP2) have been derived (step S3203). When all three control points (CP0, CP1, CP2) are derived (step S3203: YES), an affine model using the three control points (CP0, CP1, CP2) is used as an affine construction predicted motion vector (step). S3204). When the two control point mode is selected without selecting the three control point mode (step S3202: NO), it is determined whether or not all the two control points (CP0 and CP1) have been derived (step S3205). When all two control points (CP0, CP1) are derived (step S3205: YES), an affine model using the two control points (CP0, CP1) is used as an affine construction predicted motion vector (step S3206).

<Derivation of affine same predicted motion vector candidates>
The affine same prediction motion vector candidate derivation unit 363 will be described. The affine same prediction motion vector candidate derivation unit 463 is the same as the affine same prediction motion vector candidate derivation unit 363.

The affine same predicted motion vector candidate can be obtained by deriving the same motion vector at each control point.

Specifically, it is obtained by deriving each control point information and setting all the control points to be the same in any of CP0 to CP2, similarly to the affine construction prediction motion vector candidate derivation unit 362 and 462. It can also be obtained by setting the derived time motion vector at all control points in the same manner as in the normal predicted motion vector mode.

<Derivation of subblock merge mode>
Subblock merge mode derivation will be described.

FIG. 28 is a block diagram of the sub-block merge mode derivation unit 304 in the coding apparatus of this embodiment. The subblock merge mode derivation unit 304 includes a subblock merge candidate list subblockMergeCandList. This has a list structure similar to the merge candidate list mergeCandList in the normal merge mode derivation unit 302, and stores the merge index indicating the location inside the subblock merge candidate list and the subblock merge candidates corresponding to the indexes as elements. A storage area is provided. In the present embodiment, the subblockMergeCandList can register at least five merge candidates (inter-prediction information). However, each merge candidate further has motion vector information in subblock units, or has motion vector information of control points.

First, the sub-block time merge candidate derivation unit 381 derives the sub-block time merge candidate. The details of deriving the subblock time merge candidate will be described later.

Subsequently, the affine inheritance merge candidate derivation unit 382 derives the affine inheritance merge candidate. Details of affine inheritance merge candidate derivation will be described later.

Subsequently, the affine construction merge candidate derivation unit 383 derives the affine construction merge candidate. Details of deriving affine construction merge candidates will be described later.

Subsequently, the affine fixed merge candidate derivation unit 385 derives the affine fixed merge candidate. Details of deriving affine fixed merge candidates will be described later.

The subblock merge candidate selection unit 386 is a subblock merge candidate derived from the subblock time merge candidate derivation unit 381, the affine inheritance merge candidate derivation unit 382, the affine construction merge candidate derivation unit 383, and the affine fixed merge candidate derivation unit 385. A sub-block merge candidate is selected from the list, and information about the selected sub-block merge candidate is supplied to the inter-prediction mode determination unit 305.

FIG. 29 is a block diagram of the sub-block merge mode derivation unit 404 in the decoding device of the present embodiment. The subblock merge mode derivation unit 404 includes a subblock merge candidate list subblockMergeCandList. This is the same as the subblock merge mode derivation unit 304.

First, the sub-block time merge candidate derivation unit 481 derives the sub-block time merge candidate. The processing of the sub-block time merge candidate derivation unit 481 is the same as the processing of the sub-block time merge candidate derivation unit 381.

Subsequently, the affine inheritance merge candidate derivation unit 482 derives the affine inheritance merge candidate. The processing of the affine inheritance merge candidate derivation unit 482 is the same as the processing of the affine inheritance merge candidate derivation unit 382.

Subsequently, the affine construction merge candidate derivation unit 483 derives the affine construction merge candidate. The processing of the affine construction merge candidate derivation unit 483 is the same as the processing of the affine construction merge candidate derivation unit 383.

Subsequently, the affine fixed merge candidate derivation unit 485 derives the affine fixed merge candidate. The processing of the affine fixed merge candidate derivation unit 485 is the same as the processing of the affine fixed merge candidate derivation unit 485.

The subblock merge candidate selection unit 486 is a subblock merge candidate derived from the subblock time merge candidate derivation unit 481, the affine inheritance merge candidate derivation unit 482, the affine construction merge candidate derivation unit 483, and the affine fixed merge candidate derivation unit 485. A sub-block merge candidate is selected based on the index transmitted and decoded from the encoding device, and information about the selected sub-block merge candidate is supplied to the motion compensation prediction unit 406.

In the subblock merge mode derivation unit 304 and the subblock merge mode derivation unit 404, when the size (product of width and height) of a certain coded block is less than 32, the subblock merge candidate is selected in the parent block of the coded block. Derived. Then, in all the child blocks, the sub-block merge candidates derived in the parent block are used. However, this is limited to cases where the size of the parent block is 32 or more and fits within the screen.

<Derivation of subblock time merge candidates>
The operation of the sub-block time merge candidate derivation unit 381 will be described later.

<Derivation of affine inheritance merge candidates>
The affine inheritance merge candidate derivation unit 382 will be described. The affine inheritance merge candidate derivation unit 482 is the same as the affine inheritance merge candidate derivation unit 382.

Affine inheritance The merge candidate inherits the affine model of the control point from the affine model of spatially adjacent blocks.

FIG. 32 is a diagram illustrating affine inheritance merge candidate derivation. The derivation of the affine inheritance inheritance merge mode candidate can be obtained by searching the motion vector of the control point of the spatially adjacent coded / decoded blocks in the same manner as the derivation of the affine inheritance prediction motion vector.

Specifically, a maximum of one affine mode is searched for from the blocks (A0, A1) adjacent to the left side of the block to be processed and the blocks (B0, B1, B2) adjacent to the upper side of the block to be processed. , Used for affine merge mode.

FIG. 36 is a flowchart of affine inheritance merge candidate derivation.

First, the blocks (A0, A1) adjacent to the left side of the block to be processed are set as the left group (step S3301), and it is determined whether or not the block including A0 is in the affine mode (step S3302). When A0 is in the affine mode (step S3102: YES), the affine model used by A0 is acquired (step S3303), and the process proceeds to the processing of the block adjacent to the upper side. When A0 is not in affine mode (step S3302: NO), the target of deriving the affine inheritance merge candidate is set to A0-> A1, and acquisition of affine mode is attempted from the block including A1.

Subsequently, the blocks (B0, B1, B2) adjacent to the upper side of the block to be processed are set as the upper group (step S3304), and it is determined whether or not the block including B0 is in the affine mode (step S3305). When B0 is in the affine mode (step S3305: YES), the affine model used by B0 is acquired (step S3306), and the process ends. When B0 is not in affine mode (step S3305: NO), the target of deriving the affine inheritance merge candidate is set to B0-> B1, and acquisition of affine mode is attempted from the block including B1. Further, when B1 is not in the affine mode (step S3305: NO), the target of deriving the affine inheritance merge candidate is set to B1-> B2, and the acquisition of the affine mode is attempted from the block including B2.

<Derivation of affine construction merge candidates>
The affine construction merge candidate derivation unit 383 will be described. The affine construction merge candidate derivation unit 483 is the same as the affine construction merge candidate derivation unit 383.

FIG. 33 is a diagram illustrating the derivation of affine construction merge candidates. The affine construction merge candidate constructs an affine model of a control point from the motion information and time-coded blocks of spatially adjacent blocks.

Specifically, the motion vector of the upper left control point CP0 is derived from the blocks (B2, B3, A2) adjacent to the upper left side of the block to be processed, and the blocks (B1, B0) adjacent to the upper right side of the block to be processed are derived. ), The motion vector of the upper right control point CP1 is derived, and the motion vector of the lower left control point CP2 is derived from the blocks (A1, A0) adjacent to the lower left side of the block to be processed, and the lower right side of the block to be processed. The motion vector of the lower right control point CP3 is derived from the coding block (T0) adjacent to.

FIG. 37 is a flowchart of deriving affine construction merge candidates.

First, the upper left control point CP0, the upper right control point CP1, the lower left control point CP2, and the lower right control point CP3 are derived (step S3401). The upper left control point CP0 is calculated by searching for blocks having motion information in the order of priority of B2, B3, and A2 blocks. The upper right control point CP1 is calculated by searching for blocks having motion information in the order of priority of B1 and B0 blocks. The lower left control point CP2 is calculated by searching for blocks having motion information in the order of priority of A1 and A0 blocks. The lower right control point CP3 is calculated by searching for motion information of a time block.

Subsequently, it is determined whether or not the affine model with three control points can be constructed by the derived CP0, CP1, and CP2 (step S3402), and if it can be constructed (step S3402: YES), CP0 and CP1 , CP2 three control point affine model is used as an affine merge candidate (step S3403).

Subsequently, it is determined whether or not an affine model with three control points can be constructed by the derived CP0, CP1, and CP3 (step S3404), and if it can be constructed (step S3404: YES), CP0 and CP1 , CP3 three control point affine model is used as an affine merge candidate (step S3405).

Subsequently, it is determined whether or not an affine model with three control points can be constructed by the derived CP0, CP2, and CP3 (step S3406), and if it can be constructed (step S3406: YES), CP0 and CP2 , CP3 three control point affine model is used as an affine merge candidate (step S3407).

Subsequently, it is determined whether or not the affine model with three control points can be constructed by the derived CP1, CP2, and CP3 (step S3408), and if it can be constructed (step S3408: YES), CP1, CP2 , CP3 three control point affine model is used as an affine merge candidate (step S3409).

Subsequently, it is determined whether or not the affine model with two control points can be constructed by the derived CP0 and CP1 (step S3410), and if it can be constructed (step S3410: YES), 2 by CP0 and CP1. This control point affine model is used as an affine merge candidate (step S3411).

Subsequently, it is determined whether or not an affine model with two control points can be constructed by the derived CP0 and CP2 (step S3412), and if it can be constructed (step S3412: YES), 2 by CP0 and CP2. This control point affine model is used as an affine merge candidate (step S3413).

Here, whether or not the affine model can be constructed is conditioned on at least that the reference images of all the control points are the same (affine transformation is possible). Further, the affine models other than the three control point affine models by CP0, CP1 and CP2 and the two control point affine models by CP0 and CP1 are the three control point affine models by CP0, CP1 and CP2. The two-control affine model is converted into a two-control point affine model by CP0 and CP1.

<Derivation of affine fixed merge candidates>
The affine fixed merge candidate derivation unit 385 will be described. The affine fixed merge candidate deriving unit 485 is the same as the affine fixed merging candidate deriving unit 385.

The affine fixed merge candidate fixes the motion information of the control point with the fixed motion information.

Specifically, the motion vector of each control point is fixed at (0,0).

<Derivation of time prediction motion vector>
Prior to the explanation of the time prediction motion vector, the temporal context of the picture will be described with reference to FIG. 49. FIG. 49A shows the relationship between the coded block to be processed and the coded picture that is different in time from the picture to be processed. A specific encoded picture to be referred to in encoding the picture to be processed is defined as ColPic. ColPic is identified by syntax.

Further, FIG. 49 (b) shows a coded coded block existing at the same position as the coded block to be processed and in the vicinity thereof in ColPic. However, the coding blocks of T0 and T1 shown in FIG. 49B are schematic, and the actual positions and sizes are not limited to this. Now, for the coded block to be processed, the position is (xCb, yCb), the width is cbWidth, and the height is cbHeight. And
xColBr = xCb + cbWidth
yColBr = yCb + cbHeight
Is calculated. The coded block on ColPic containing the position ((xColBr >> 3) << 3, (yColBr >> 3) << 3) is T0. Also,
xColCtr = xCb + (cbWidth >> 1)
yColCtr = yCb + (cbHeight >> 1)
Is calculated. The coded block on ColPic containing the position ((xColCtr >> 3) << 3, (yColCtr >> 3) << 3) is T1.

The above description of the temporal context of the picture is that at the time of encoding, but the same applies at the time of decoding. That is, at the time of decoding, the coding in the above description is replaced with decoding, and the same description is given.

The operation of the time prediction motion vector candidate derivation unit 322 in the normal prediction motion vector mode derivation unit 301 of FIG. 17 will be described with reference to FIG. 50.

First, ColPic is derived (step S4201). The derivation of ColPic will be described with reference to FIG.

When the slice type slice_type is B slice and the flag collapse_from_l0_flag is 0 (step S4211: YES, step S4212: YES), the picture ColPic at different times becomes the picture RefPicList1 [0] whose reference index of the reference list L1 is 0 (step S4211: YES, step S4212: YES). Step S4213). Otherwise, that is, if the slice type slice_type is a B slice and the flag collapse_from_l0_flag is 1 (step S4211: YES, step S4212: NO), or if the slice type slice_type is a P slice (step S4211: NO, step S4214: YES), the picture ColPic at different times becomes the picture RefPicList0 [0] whose reference index of the reference list L0 is 0 (step S4215). If slice_type is not a P slice (step S4214: NO), the process ends.

See FIG. 50 again. After deriving ColPic, the coding block colCb is derived and the coding information is acquired (step S4202). This process will be described with reference to FIG.

First, in the picture ColPic at different times, the coded block including the lower right position at the same position as the coded block to be coded is designated as the coded block colCb at different times (step S4221). An example of this coded block is shown in the coded block T0 of FIG. 49.

Next, the coding information of the coding block colCb at different times is acquired (step S4222). If the PredMode of the coded blocks colCb at different times is not available, or the prediction mode PredMode of the coded blocks colCb at different times is intra-prediction (MODE_INTRA) (step S4223: NO, step S4224: YES), pictures at different times. A coded block including the lower right center position at the same position as the coded block to be processed in ColPic is set as a coded block colCb at a different time (step S4225). An example of this coded block is shown in the coded block T1 of FIG.

See FIG. 50 again. Next, inter-prediction information is derived for each reference list (steps S4203 and S4204). Here, for the coding block colCb, the motion vector mvLXCol for each reference list and the flag availableFlagLXCol indicating whether or not the coding information is valid are derived. LX indicates a reference list, LX becomes L0 in the derivation of reference list 0, and LX becomes L1 in the derivation of reference list 1. The derivation of the inter-prediction information will be described with reference to FIG. 53.

When the coding blocks colCb at different times are not available (step S4231: NO), or when the prediction mode PredMode is intra-prediction (MODE_INTRA) (step S4232: NO), both the flag availableFlagLXCol and the flag predFlagLXCol are set to 0 (step S4233). , The motion vector mvLXCol is set to (0,0) (step S4234), and the process ends.

When the coding block colCb is available (step S4231: YES) and the prediction mode PredMode is not intra-prediction (MODE_INTRA) (step S4232: YES), mvCol, refIdxCol and availableFlagCol are calculated by the following procedure.

When the flag PredFlagL0 [xPCol] [yPCol] indicating whether or not the L0 prediction of the coded block colCb is used is 0 (step S4235: YES), the prediction mode of the coded block colCb is Pred_L1, so the motion vector. mvCol is set to the same value as MvL1 [xPCol] [yPCol] which is the motion vector of L1 of the coding block colCb (step S4236), and the reference index refIdxCol is set to the same value as the reference index RefIdxL1 [xPCol] [yPCol] of L1. It is set (step S4237), and the reference list listCol is set to L1 (step S4238). Here, xPCol and yPCol are indexes indicating the positions of the upper left pixels of the coding block colCb in the picture ColPic at different times.

On the other hand, when the L0 prediction flag PredFlagL0 [xPCol] [yPCol] of the coding block colCb is not 0 (step S4235: NO), it is determined whether or not the L1 prediction flag PredFlagL1 [xPCol] [yPCol] of the coding block colCb is 0. To do. When the L1 prediction flag PredFlag L1 [xPCol] [yPCol] of the coding block colCb is 0 (step S4239: YES), the motion vector mvCol is the same as MvL0 [xPCol] [yPCol] which is the motion vector of L0 of the coding block colCb. It is set to a value (step S4240), the reference index refIdxCol is set to the same value as the reference index RefIdxL0 [xPCol] [yPCol] of L0 (step S4241), and the reference list listCol is set to L0 (step S4242).

If the L0 prediction flag PredFlag L0 [xPCol] [yPCol] of the coding block colCb and the L1 prediction flag PredFlag L1 [xPCol] [yPCol] of the coding block colCb are both non-zero (step S4235: NO and S4239: NO), the coding Since the inter-prediction mode of the block colCb is bi-prediction (Pred_BI), one is selected from the two motion vectors L0 and L1 (step S4243).

FIG. 54 is a flowchart showing a procedure for deriving the inter-prediction information of the coded block when the inter-prediction mode of the coded block colCb is bi-prediction (Pred_BI).

First, it is determined whether or not the POCs of all the pictures registered in all the reference lists are smaller than the POCs of the currently processed pictures (step S4251), and L0 and all the reference lists of the coding block colCb are used. When the POC of all the pictures registered in L1 is smaller than the POC of the current picture to be processed (step S4251: YES), LX is L0, that is, a prediction vector candidate for the motion vector of L0 of the coded block to be processed. (Step S4252: YES), the inter-prediction information of L0 of the coding block colCb is selected, and LX selects the prediction vector candidate of the motion vector of L1 of the coding block to be processed. When deriving (step S4252: NO), the inter-prediction information of L1 of the coding block colCb is selected. On the other hand, when at least one of the POCs of the pictures registered in all the reference lists L0 and L1 of the coding block colCb is larger than the POC of the currently processed picture (step S4251: NO), and the flag collapse_from_l0_flag is 0 (step S4251: NO). Step S4253: YES), the inter-prediction information for L0 of the coding block colCb is selected, and when the flag collocated_from_l0_flag is 1 (step S4253: NO), the inter-prediction information for L1 of the coding block colCb is selected. ..

When the inter-prediction information toward L0 of the coding block colCb is selected (step S4252: YES or step S4253: YES), the motion vector mvCol is set to the same value as MvL0 [xPCol] [yPCol] (step S4254). , The reference index refIdxCol is set to the same value as RefIdxL0 [xPCol] [yPCol] (step S4255), and the list listCol is set to L0 (step S4256).

When the inter-prediction information for L1 of the coding block colCb is selected (step S4252: NO or step S4253: NO), the motion vector mvCol is set to the same value as MvL1 [xPCol] [yPCol] (step S4257). , The reference index refIdxCol is set to the same value as RefIdxL1 [xPCol] [yPCol] (step S4258), and the list listCol is set to L1 (step S4259).

Returning to FIG. 53, when the inter-prediction information can be obtained from the coding block colCb, both the flag availableFlagLXCol and the flag predFlagLXCol are set to 1 (step S4244).

Subsequently, the motion vector mvCol is scaled to obtain the motion vector mvLXCol (step S4245). The scaling calculation processing procedure of this motion vector mvLXCol will be described with reference to FIG. 55.

From the POC of the picture ColPic at different times, subtract the POC of the reference picture corresponding to the reference index refIdxCol referenced in the list listCol of the coding block colCb to obtain the inter-picture distance td.
td = [POC of picture ColPic at different times]-[POC of reference picture referenced in list Col of coding block colCb]
Is calculated (step S4261). If the POC of the reference picture referenced in the list listCol of the coded block colCb is earlier in the display order than the picture ColPic at different times, the inter-picture distance td becomes a positive value and is larger than the picture ColPic at different times. If the POC of the reference picture referenced in the list listCol of the coding block colCb is later in the display order, the inter-picture distance td is a negative value.

Next, the distance tb between the pictures is calculated by subtracting the POC of the reference picture referenced by the list LX of the current processing target pictures from the POC of the current processing target picture.
tb = [POC of the currently processed picture]-[POC of the reference picture corresponding to the reference index of the LX of the time merge candidate]
Is calculated (step S4262). If the reference picture referred to in the list LX of the current processing target pictures is earlier in the display order than the current processing target picture, the inter-picture distance tb becomes a positive value, and the list LX of the current processing target pictures. If the reference pictures referred to in are later in the display order, the inter-picture distance tb has a negative value.

Subsequently, the inter-picture distances td and tb are compared (step S4263), and when the inter-picture distances td and tb are equal (step S4263: YES), the motion vector mvLXCol is set.
mvLXCol = mvCol
(Step S4264), and the current scaling calculation process is completed.

On the other hand, when the inter-picture distances td and tb are not equal (step S4263: NO), the variable tx is set to
tx = (16384 + Abs (td) >> 1) / td
Is calculated (step S4265). Next, the scaling coefficient distScaleFactor,
distScaleFactor = Clip3 (-4096, 4095, (tb * tx + 32) >> 6)
Is calculated (step S4266). Here, Clip3 (x, y, z) is a function that limits the minimum value to x and the maximum value to y for the value z. Next, the motion vector mvLXCol,
mvLXCol = Clip3 (-32768, 32767, Sign (distScaleFactor * mvLXCol))
* ((Abs (distScaleFactor * mvLXCol) + 127) >> 8))
(Step S4267), and the current scaling calculation process is completed. Here, Sign (x) is a function that returns the sign of the value x, and Abs (x) is a function that returns the absolute value of the value x.

See FIG. 50 again. Then, the motion vector mvL0Col of L0 is added as a candidate to the predicted motion vector candidate list mvpListLX in the above-mentioned normal predicted motion vector mode derivation unit 301 (step S4205). However, this addition is made only when the flag availableFlagL0Col = 1, which indicates whether the coding block colCb in reference list 0 is valid. Further, the motion vector mvL1Col of L1 is added as a candidate to the predicted motion vector candidate list mvpListLX in the above-mentioned normal predicted motion vector mode derivation unit 301 (step S4205). However, this addition is made only when the flag availableFlagL1Col = 1, which indicates whether the encoding block colCb in the reference list 1 is valid. As described above, the processing of the time prediction motion vector candidate derivation unit 322 is completed.

The above description of the normal prediction motion vector mode derivation unit 301 is for coding, but the same applies for decoding. That is, the operation of the time prediction motion vector candidate derivation unit 422 in the normal prediction motion vector mode derivation unit 401 of FIG. 23 is similarly described by replacing the coding in the above description with decoding.

<Derivation of time merge candidates>
The operation of the time merge candidate derivation unit 342 in the normal merge mode derivation unit 302 of FIG. 18 will be described with reference to FIG. 56.

First, ColPic is derived (step S4301). Next, the coding block colCb is derived and the coding information is acquired (step S4302). Further, inter-prediction information is derived for each reference list (steps S4303 and S4304). Since the above processing is the same as S4201 to S4204 in the time prediction motion vector candidate derivation unit 322, the description thereof will be omitted.

Next, the flag availableFlagCol indicating whether or not the coding block colCb is valid is calculated (step S4305). When the flag availableFlagL0Col or the flag availableFlagL1Col is 1, the availableFlagCol is 1. Otherwise, availableFlagCol is 0.

Then, the motion vector mvL0Col of L0 and the motion vector mvL1Col of L1 are added as candidates to the merge candidate list mergeCandList in the above-mentioned normal merge mode derivation unit 302 (step S4306). However, this addition is only when the flag availableFlagCol = 1, which indicates whether the coding block colCb is valid or not. As described above, the processing of the time merge candidate derivation unit 342 is completed.

The above description of the time merge candidate derivation unit 342 is for encoding, but the same applies for decoding. That is, the operation of the time merge candidate derivation unit 442 in the normal merge mode derivation unit 402 of FIG. 24 is similarly described by replacing the coding in the above description with decoding.

<Update of history prediction motion vector candidate list>
Next, a method of initializing and updating the history prediction motion vector candidate list HmvpCandList provided in the coding information storage memory 111 on the coding side and the coding information storage memory 205 on the decoding side will be described in detail. FIG. 38 is a flowchart illustrating a procedure for initializing / updating the history prediction motion vector candidate list.

In the present embodiment, the history prediction motion vector candidate list HmvpCandList is updated by the coded information storage memory 111 and the coded information storage memory 205. A history candidate list update unit may be installed in the inter-prediction unit 102 and the inter-prediction unit 203 to update the history prediction motion vector candidate list HmvpCandList.

Initialize the history prediction motion vector candidate list HmvpCandList at the beginning of the slice, and update the history prediction motion vector candidate list HmvpCandList on the coding side when the normal prediction vector mode or the normal merge mode is selected in the prediction method determination unit 105. Then, on the decoding side, the history prediction motion vector candidate list HmvpCandList is updated when the inter-prediction mode decoded by the bit string decoding unit 201 is the normal prediction vector mode or the normal merge mode.

The inter-prediction information used when performing inter-prediction in the normal prediction vector mode or the normal merge mode is registered in the historical prediction motion vector candidate list HmvpCandList as the inter-prediction information candidate hMvpCand. The inter-prediction information candidate hMvpCand includes the reference index refIdxL0 of L0 and the reference index refIdxL1 of L1, the L0 prediction flag predFlag L0 indicating whether L0 prediction is performed, and the L1 prediction flag predFlag L1 indicating whether L1 prediction is performed. The motion vector mvL0 of L0 and the motion vector mvL1 of L1 are included. Inter-prediction information candidate among the elements (that is, inter-prediction information) registered in the history prediction motion vector candidate list HmvpCandList provided in the coding information storage memory 111 on the coding side and the coding information storage memory 205 on the decoding side. If the inter-prediction information with the same value as hMvpCand exists, delete the element from the historical prediction motion vector candidate list HmvpCandList. On the other hand, if there is no inter-prediction information with the same value as the inter-prediction information candidate hMvpCand, the first element of the historical prediction motion vector candidate list HmvpCandList is deleted, and the inter-prediction information candidate is at the end of the historical prediction motion vector candidate list HmvpCandList. Add hMvpCand.

The number of elements of the history prediction motion vector candidate list HmvpCandList provided in the coding information storage memory 111 on the coding side and the coding information storage memory 205 on the decoding side of the present invention is 6.

First, the history prediction motion vector candidate list HmvpCandList for each slice is initialized (step S2101 in FIG. 38). Empty all the elements of the history prediction motion vector candidate list HmvpCandList at the beginning of the slice, and set the value of the number of history prediction motion vector candidates registered in the history prediction motion vector candidate list HmvpCandList to 0.

Although the history prediction motion vector candidate list HmvpCandList is initialized in slice units (the first coded block of the slice), it may be performed in picture units, tile units, or tree block line units.

Subsequently, the following history prediction motion vector candidate list HmvpCandList update process is repeated for each coded block in the slice (steps S2102 to S2107 in FIG. 38).

First, the initial setting is performed for each coded block. A FALSE (false) value is set in the flag identicalCandExist indicating whether or not the same candidate exists, and 0 is set in the removal target index removeIdx (step S2103 in FIG. 38).

It is determined whether or not the inter-prediction information candidate hMvpCand to be registered exists in the history prediction motion vector candidate list HmvpCandList (step S2104 in FIG. 38). When the prediction method determination unit 105 on the coding side determines the normal prediction motion vector mode or the normal merge mode, or when the bit string decoding unit 201 on the decoding side decodes the decoding as the normal prediction motion vector mode or the normal merge mode. Let the inter-prediction mode be hMvpCand. When the coding side prediction method determination unit 105 determines the intra prediction mode, subblock prediction motion vector mode or subblock merge mode, or the decoding side bit string decoding unit 201 determines the intra prediction mode, subblock prediction motion vector mode. Or, when decoded in the sub-block merge mode, the history prediction motion vector candidate list HmvpCandList is not updated, and the inter-prediction information candidate hMvpCand to be registered does not exist. If the inter-prediction information candidate hMvpCand to be registered does not exist, steps S2105 to S2106 are skipped (step S2104: NO in FIG. 38). If the inter-prediction information candidate hMvpCand to be registered exists, the processing of step S2105 or less is performed (step S2104: YES in FIG. 38).

Subsequently, it is determined whether or not the same element as the inter-prediction information candidate hMvpCand to be registered exists in each element of the history prediction motion vector candidate list HmvpCandList (step S2105 in FIG. 38). FIG. 39 is a flowchart of the same element confirmation processing procedure. Number of history prediction motion vector candidates When the value of NumHmvpCand is 0 (step S2121: NO in FIG. 39), the history prediction motion vector candidate list HmvpCandList is empty and the same candidate does not exist, so skip steps S212 to S2125 in FIG. 39. , This same element confirmation processing procedure is terminated. When the number of historically predicted motion vector candidates NumHmvpCand is greater than 0 (step S2121: YES in FIG. 39), the process of step S2123 is repeated from 0 to NumHmvpCand-1 in the historical predicted motion vector index hMvpIdx (step in FIG. 39). S2122-S2125). First, it is compared whether or not the hMvpCandList [hMvpIdx], which is the xth element of the historical prediction motion vector candidate list from 0, is the same as the inter-prediction information candidate hMvpCand (step S2123 in FIG. 39). If they are the same (step S2123: YES in FIG. 39), a TRUE (true) value is set in the flag identicalCandExist indicating whether or not the same candidate exists, and a value of hMVpIndex is set in the removal target index removeIdx. The element confirmation process ends. If they are not the same (step S2123: NO in FIG. 39), hMvpIdx is incremented by 1, and if the historical prediction motion vector index hMvpIdx is NumHmvpCand-1 or less, the processes after step S2123 are performed (steps S212 to S2125 in FIG. 39). ..

Returning to the flowchart of FIG. 38 again, the elements of the history prediction motion vector candidate list HmvpCandList are shifted and added (step S2106 of FIG. 38). FIG. 40 is a flowchart of the element shift / addition processing procedure of the history prediction motion vector candidate list HmvpCandList in step S2106 of FIG. 38. First, it is determined whether to remove the elements stored in the historical prediction motion vector candidate list HmvpCandList and then add a new element, or to add a new element without removing the elements. Specifically, it is compared whether or not the flag identicalCandExist indicating whether or not the same candidate exists is TRUE (true) or the NumHmvpCand is 6 (step S2141 in FIG. 40). If the flag identicalCandExist indicating whether or not the same candidate exists satisfies either TRUE (true) or NumHmvpCand satisfies either of 6 (step S2141: YES in FIG. 40), it is stored in the historical prediction motion vector candidate list HmvpCandList. Remove the existing element and then add a new element. Set the initial value of index i to the value of removeIdx + 1. The element shift process in step S2143 is repeated from this initial value to NumHmvpCand. (Steps S2142 to S2144 in FIG. 40). By copying the element of HMVPCandList [i] to HMVPCandList [i-1], the element is shifted forward (step S2143 in FIG. 40) and i is incremented by 1 (step S2142 to S2144 in FIG. 40). When the index i becomes NumHmvpCand + 1 and the element shift processing in step S2143 is completed, the inter-prediction information candidate hMvpCand is added to the end of the history prediction motion vector candidate list (step S2145 in FIG. 40). Here, the end of the history prediction motion vector candidate list is the (NumHmvpCand-1) th HMVPCandList [NumHmvpCand-1] counting from 0. This completes the element shift / addition process of this history prediction motion vector candidate list HMVPCandList. On the other hand, when the flag identicalCandExist indicating whether or not the same candidate exists does not satisfy any of the conditions of TRUE (true) and NumHmvpCand of 6 (step S2141: NO in FIG. 40), it is stored in the historical prediction motion vector candidate list HmvpCandList. The inter-prediction information candidate hMvpCand is added to the end of the history prediction motion vector candidate list without excluding the elements (step S2146 in FIG. 40). Here, the last of the history prediction motion vector candidate list is the HMV PCandList [NumHmvpCand] which is the NumHmvpCand th from 0. In addition, NuMHmvpCand is incremented by 1 to end the element shift / addition process of this history prediction motion vector candidate list HMVPCandList.

FIG. 43 is a diagram illustrating an example of the update process of the history prediction motion vector list. When 6 elements (inter prediction information) are registered in the history prediction motion vector candidate list HMVPCandList and new inter prediction information is added, each element of the history prediction motion vector candidate list HMVPCandList and a new inter from the front Comparing the prediction information (Fig. 43 (a)), if the new inter-prediction information has the same value as the third element from the beginning of the historical prediction motion vector candidate list HMVPCandList, HMVP2, from the historical prediction motion vector candidate list HMVPCandList. Delete the element HMVP2, shift (copy) the rear elements HMVP3 to HMVP5 one by one, and add new inter-prediction information to the end of the history prediction motion vector candidate list HMVPCandList (Fig. 43 (b)). , The update of the history prediction motion vector candidate list HMVPCandList is completed (FIG. 43 (c)).

<History prediction motion vector candidate derivation process>
Next, the history prediction motion vector candidate derivation unit 323 of the normal prediction motion vector mode derivation unit 301 on the coding side and the history prediction motion vector candidate derivation unit 423 of the normal prediction motion vector mode derivation unit 401 on the decoding side perform common processing. A method of deriving the history prediction motion vector candidate from the history prediction motion vector candidate list HMVPCandList, which is the processing procedure of step S304 of FIG. 20, will be described in detail. FIG. 41 is a flowchart illustrating a history prediction motion vector candidate derivation processing procedure.

Current number of predicted motion vector candidates numCurrMvpCand is greater than or equal to the maximum number of elements in the predicted motion vector candidate list mvpListLX (here, 2) or the number of historical predicted motion vector candidates When the value of NumHmvpCand is 0 (step S2201: in FIG. 41) NO), the process of steps S2202 to S2209 of FIG. 41 is omitted, and the history prediction motion vector candidate derivation process procedure is completed. When the current number of predicted motion vector candidates numCurrMvpCand is less than 2, which is the maximum number of elements in the predicted motion vector candidate list mvpListLX, and when the value of the number of historical predicted motion vector candidates NumHmvpCand is greater than 0 (step S2201: in FIG. 41). YES), the processes of steps S2202 to S2209 of FIG. 41 are performed.

Subsequently, the processes of steps S2203 to S2208 of FIG. 41 are repeated until the index i is 0 to 3 and the number of historical prediction motion vector candidates NuMHmvpCand, whichever is smaller (steps S2202 to S2209 of FIG. 41). When the current number of predicted motion vector candidates numCurrMvpCand is 2 or more, which is the maximum number of elements of the predicted motion vector candidate list mvpListLX (step S2203: NO in FIG. 41), the processing of steps S2204 to S2209 in FIG. 41 is omitted. The history prediction motion vector candidate derivation processing procedure ends. When the current number of predicted motion vector candidates numCurrMvpCand is smaller than 2, which is the maximum number of elements of the predicted motion vector candidate list mvpListLX (step S2203: YES in FIG. 41), the processes after step S2204 in FIG. 41 are performed.

Subsequently, the processes from steps S2205 to S2207 are performed for Y's 0 and 1 (L0 and L1), respectively (steps S2204 to S2208 in FIG. 41). When the current number of predicted motion vector candidates numCurrMvpCand is 2 or more, which is the maximum number of elements of the predicted motion vector candidate list mvpListLX (steps S2205: NO in FIG. 41), the processing of steps S2206 to S2209 in FIG. 41 is omitted. The history prediction motion vector candidate derivation processing procedure ends. When the current number of predicted motion vector candidates numCurrMvpCand is smaller than 2, which is the maximum number of elements of the predicted motion vector candidate list mvpListLX (step S2205: YES in FIG. 41), the processes after step S2206 in FIG. 41 are performed.

Next, when the reference index of LY in the historical prediction motion vector candidate list HmvpCandList [i] is the same as the reference index refIdxLX of the encoded / decoded motion vector (step S2206: YES in FIG. 41), the predicted motion vector candidate list As the last element, the LY motion vector of the historical prediction motion vector candidate HmvpCandList [i] is added to the numCurrMvpCand th element mvpListLX [numCurrMvpCand] counting from 0 in the predicted motion vector candidate list (step S2207 in FIG. 41). The number of predicted motion vector candidates in numCurrMvpCand is incremented by 1. If the reference index of LY in the historical prediction motion vector candidate list HmvpCandList [i] is not the same as the reference index refIdxLX of the motion vector to be encoded / decoded (step S2206: NO in FIG. 41), the additional processing in step S2207 is skipped. ..

The above processes of steps S2205 to S2207 of FIG. 41 are performed at both L0 and L1 (steps S2204 to S2208 of FIG. 41).

When the index i is incremented by 1 and the index i is equal to or less than the smaller value of 3 and the number of historical prediction motion vector candidates NUMHmvpCand, the processing of steps S2203 and subsequent steps is performed again (steps S2202 to S2209 in FIG. 41).

<History merge candidate derivation process>
Next, the process of step S404 of FIG. 21, which is a process common to the history merge candidate derivation unit 345 of the normal merge mode derivation unit 302 on the coding side and the history merge candidate derivation unit 445 of the normal merge mode derivation unit 402 on the decoding side. The method of deriving the history merge candidate from the history merge candidate list HmvpCandList, which is a procedure, will be described in detail. FIG. 42 is a flowchart illustrating a history merge candidate derivation processing procedure.

First, the initialization process is performed (step S2301 in FIG. 42). Set the value of FALSE to each element from 0 to (numCurrMergeCand -1) of isPruned [i], and set the variable numOrigMergeCand to the number of elements registered in the current merge candidate list, numCurrMergeCand.

Subsequently, the initial value of the index hMvpIdx is set to 1, and the additional processing from step S2303 to step S2310 in FIG. 42 is repeated from this initial value to NuMHmvpCand (steps S2302 to S2311 in FIG. 42). If the number of elements registered in the current merge candidate list numCurrMergeCand is not less than or equal to (maximum number of merge candidates MaxNumMergeCand-1), merge candidates have been added to all elements in the merge candidate list, so this history merge candidate derivation The process ends (step S2303: NO in FIG. 42). When the number of elements registered in the current merge candidate list numCurrMergeCand is (maximum number of merge candidates MaxNumMergeCand-1) or less (step S2303: YES in FIG. 42), the processes after step S2304 are performed.

First, a FALSE (false) value is set in sameMotion (step S2304 in FIG. 42). Subsequently, the initial value of the index i is set to 0, and the processes of steps S2306 and S2307 of FIG. 42 are performed from this initial value to 1 (S2305 to S2308 of FIG. 42).

Next, is the (NumHmvpCand-hMvpIdx) th element HmvpCandList [NumHmvpCand-hMvpIdx] counted from 0 in the historical motion vector prediction candidate list and the i-th element mergeCandList [i] counted from 0 in the merge candidate list the same value? Whether or not they are compared (step S2306 in FIG. 42). Here, the same value of the merge candidate indicates that the values of all the components (inter-prediction mode, reference index, motion vector) of the merge candidate are the same. However, the processing in step S2306 is limited to the case where hMvpIdx is larger than NuMHmvpCand-2, mergeCandList [i] is a spatial merge candidate, and isPruned [i] is FALSE (false). If the values are the same (step S2306: YES in FIG. 39), TRUE (true) is set for both sameMotion and isPruned [i] (step S2307 in FIG. 42). If the values are not the same (step S2306: NO in FIG. 39), the process of step S2307 is skipped. When the iterative processing from step S2305 to step S2308 in FIG. 42 is completed, it is compared whether or not sameMotion is FALSE (false) (step S2309 in FIG. 42), and when sameMotion is FALSE (false) (step S2309 in FIG. 42: YES), add the (NumHmvpCand --hMvpIdx) th element HmvpCandList [NumHmvpCand --hMvpIdx] to the mergeCandList [numCurrMergeCand] of the numCurrMergeCand of the merge candidate list, and increment numCurrMergeCand1. Step S2310 in FIG. 42). The index hMvpIdx is incremented by 1 (step S2302 in FIG. 42), and the repetition of steps S2302 to S2311 in FIG. 42 is repeated.

When the confirmation of all the elements of the history prediction motion vector candidate list is completed or the merge candidates are added to all the elements of the merge candidate list, the derivation process of this history merge candidate is completed.

<Average merge candidate derivation process>
Next, the process of step S403 in FIG. 21, which is a process common to the average merge candidate derivation unit 344 of the normal merge mode derivation unit 302 on the coding side and the average merge candidate derivation unit 444 of the normal merge mode derivation unit 402 on the decoding side. The method of deriving the average merge candidate, which is a procedure, will be described in detail. FIG. 62 is a flowchart illustrating a procedure for deriving the average merge candidate.

First, the initialization process is performed (step S1301 in FIG. 62). Set the variable numOrigMergeCand to the number of elements registered in the current merge candidate list, numCurrMergeCand.

Then, the merge candidate list is scanned in order from the beginning, and two motion informations are determined. It is assumed that the index i = 0 indicating the first motion information and the index j = 1 indicating the second motion information. (Steps S1302 to S1303 in FIG. 62). If the number of elements registered in the current merge candidate list numCurrMergeCand is not less than or equal to (maximum number of merge candidates MaxNumMergeCand-1), merge candidates have been added to all elements in the merge candidate list, so this history merge candidate derivation The process ends (step S1304 in FIG. 62). If the number of elements registered in the current merge candidate list numCurrMergeCand is (maximum number of merge candidates MaxNumMergeCand-1) or less, the processing in step S1305 and subsequent steps is performed.

It is determined whether or not both the i-th motion information mergeCandList [i] of the merge candidate list and the j-th motion information mergeCandList [j] of the merge candidate list are invalid (step S1305 in FIG. 62), and both are invalid. In that case, move on to the next element without deriving the average merge candidates of mergeCandList [i] and mergeCandList [j]. If both mergeCandList [i] and mergeCandList [j] are not invalid, the following processing is repeated with X as 0 and 1 (steps S1306 to 1314 in FIG. 62).

It is determined whether the LX prediction of mergeCandList [i] is valid (step S1307 of FIG. 62). When the LX prediction of mergeCandList [i] is valid, it is determined whether the LX prediction of mergeCandList [j] is valid (step S1308 in FIG. 62). If the LX prediction of mergeCandList [j] is valid, that is, if both the LX prediction of mergeCandList [i] and the LX prediction of mergeCandList [j] are valid, then the motion vector of the LX prediction of mergeCandList [i] and the mergeCandList The average merge candidate of the LX prediction having the motion vector of the LX prediction obtained by averaging the movement vectors of the LX prediction of [j] and the reference index of the LX prediction of mergeCandList [i] is derived and set to the LX prediction of the averageCand, and the averageCand Enable LX prediction (step S1309 in FIG. 62). In step S1308 of FIG. 62, if the LX prediction of mergeCandList [j] is not valid, that is, if the LX prediction of mergeCandList [i] is valid and the LX prediction of mergeCandList [j] is invalid, then mergeCandList [i] The average merge candidate of the LX prediction having the motion vector and the reference index of the LX prediction is derived and set to the LX prediction of the averageCand, and the LX prediction of the averageCand is enabled (step S1310 in FIG. 62). If the LX prediction of mergeCandList [i] is not valid in step S1307 of FIG. 62, it is determined whether or not the LX prediction of mergeCandList [j] is valid (step S1311 of FIG. 62). If the LX prediction of mergeCandList [j] is valid, that is, if the LX prediction of mergeCandList [i] is invalid and the LX prediction of mergeCandList [j] is valid, then the motion vector of the LX prediction of mergeCandList [j] The average merge candidate of the LX prediction having the reference index is derived and set to the LX prediction of the averageCand, and the LX prediction of the averageCand is enabled (step S1312 in FIG. 62). In step S1311 of FIG. 62, if the LX prediction of mergeCandList [j] is not valid, that is, if both the LX prediction of mergeCandList [i] and the LX prediction of mergeCandList [j] are invalid, the LX prediction of averageCand is invalidated. (Step S1312 in FIG. 62).

Here, LX prediction is valid when the reference index refIdxLX is 0 or more, and when LX prediction is invalid, that is, it does not exist, the reference index refIdxLX is set to -1.

The average merge candidate averageCand of the L0 prediction, L1 prediction or BI prediction generated as described above is added to the mergeCandList [numCurrMergeCand] of the numCurrMergeCand th numCurrMergeCand in the merge candidate list, and the numCurrMergeCand is incremented by 1 (step S1315 in FIG. 62). This completes the process of deriving the average merge candidate.

The average merge candidate is averaged for each of the horizontal component of the motion vector and the vertical component of the motion vector.

<Derivation of subblock time merge candidates>
The operation of the subblock time merge candidate derivation unit 381 in the subblock merge mode derivation unit 304 of FIG. 16 will be described with reference to FIG.

First, it is determined whether or not the coded block has less than 8x8 pixels (step S4002).

When the coded block is less than 8x8 pixels (step S4002: YES), the flag availableFlagSbCol = 0 indicating the existence of the subblock time merge candidate is set (step S4003), and the processing of the subblock time merge candidate derivation unit is terminated. .. Here, when the temporal motion vector prediction is prohibited by the syntax, or when the subblock time merging is prohibited, the same processing as when the coded block is less than 8x8 pixels (step S4002: YES) is performed. To do.

On the other hand, when the coded block has 8x8 pixels or more (step S4002: NO), the adjacent movement information of the coded block in the coded picture is derived (step S4004).

The process of deriving the adjacent movement information of the coded block will be described with reference to FIG. 45. The process of deriving the adjacent motion information is similar to the process of the spatial prediction motion vector candidate deriving unit 321 described above. However, the order of searching adjacent blocks is A0, B0, B1, A1, and B2 is not searched. First, the coding information is acquired with the adjacent block n = A0 (step S4052). The coding information indicates a flag availableFlagN indicating whether or not adjacent blocks can be used, a reference index refIdxLXN for each reference list, and a motion vector mvLXN.

Next, it is determined whether the adjacent block n is valid or invalid (step S4054). If the flag availableFlagN = 1, which indicates whether the adjacent block can be used, is valid, otherwise it is invalid.

If the adjacent block n is valid (step S4054: YES), the reference index refIdxLXN is set to the reference index refIdxLXn of the adjacent block n (step S4056). Further, the motion vector mvLXN is set as the motion vector mvLXn of the adjacent block n (step S4056), and the process of deriving the adjacent motion information of the blocks is completed.

On the other hand, if the adjacent block n is invalid (step S4054: NO), the coding information is acquired with the adjacent block n = B0 (step S4052), and it is determined whether the adjacent block n is valid or invalid (step S4054). .. After that, the same processing is performed to loop in the order of B1 and A1. The process of deriving the adjacent motion information loops until the adjacent block becomes valid, and if all the adjacent blocks A0, B0, B1, A1 are invalid, the process of deriving the adjacent motion information of the block ends.

See FIG. 44 again. After deriving the adjacent motion information (step S4004), the temporal motion vector is derived (step S4006).

The process of deriving the temporal motion vector will be described with reference to FIG. First, it is initialized as the temporal motion vector tempMv = (0,0) (step S4062).

Next, it is determined whether the adjacent movement information is valid or invalid (step S4064). If the flag availableFlagN = 1, which indicates whether the adjacent block can be used, is valid, otherwise it is invalid. When the adjacent motion information is invalid (step S4064: NO), the process of deriving the temporal motion vector is terminated.

On the other hand, when the adjacent movement information is valid (step S4064: YES), it is determined whether or not the flag predFlagL1N indicating whether or not the L1 prediction is used in the adjacent block N is 1 (step S4066). When predFlagL1N = 0 (step S4066: NO), the process proceeds to the next process (step S4078). When predFlagL1N = 1 (step S4066: YES), it is determined whether or not the POC of all the pictures registered in all the reference lists is equal to or less than the POC of the current processing target picture (step S4068). If this determination is true (step S4068: YES), the process proceeds to the next process (step S4070).

If the slice type slice_type is B slice and the flag collapse_from_l0_flag is 0 (step S4070: YES, step and S4072: YES), ColPic and the reference picture RefPicList1 [refIdxL1N] (picture of the reference index refIdxL1N of the reference list L1) are the same. Is determined (step S4074). If this determination is true (step S4074: YES), the temporal motion vector tempMv = mvL1N is set (step S4076). If this determination is false (step S4074: NO), the process proceeds to the next process (step S4078). If the slice type slice_type is not a B slice and the flag collapse_from_l0_flag is not 0 (step S4070: NO or step S4072: NO), the process proceeds to the next process (step S4078).

Then, it is determined whether or not the flag predFlagL0N indicating whether or not the L0 prediction is used in the adjacent block N is 1 (step S4078). When predFlagL0N = 1 (step S4078: YES), it is determined whether ColPic and the reference picture RefPicList0 [refIdxL0N] (picture of the reference index refIdxL0N of the reference list L0) are the same (step S4080). If this determination is true (step S4080: YES), the temporal motion vector tempMv = mvL0N is set (step S4082). If this determination is false (step S4080: NO), the process of deriving the temporal motion vector is terminated.

See FIG. 44 again. Next, ColPic is derived (step S4016). Since this process is the same as S4201 in the time prediction motion vector candidate derivation unit 322, the description thereof will be omitted.

Then, the coding blocks colCb at different times are set (step S4017). In this method, the coded block located at the lower right of the center at the same position as the coded block to be processed in the picture ColPic at different times is set as colCb. This coded block corresponds to the coded block T1 in FIG.

Next, the position where the temporal motion vector tempMv is added to the coding block colCb is set as a new colCb (step S4018). Now, let the upper left position of the coding block colCb be (xColCb, yColCb) and the temporal motion vector tempMv with 1/16 pixel precision (tempMv [0], tempMv [1]). And
xColCb = Clip3 (xCtb, xCtb + CtbSizeY + 3, xColCb + (tempMv [0] >> 4))
yColCb = Clip3 (yCtb, yCtb + CtbSizeY --1, yColCb + (tempMv [1] >> 4))
Is calculated. Here, the upper left position of the tree block is (xCtb, yCtb), and the size of the tree block is CtbSizeY. The coded block on ColPic containing the position ((xColCb >> 3) << 3, (yColCb >> 3) << 3) becomes the new colCb. As shown in the above equation, the position after tempMv addition is corrected to a range of about the size of the tree block so that it does not deviate significantly compared to before tempMv addition. If this position is off the screen, it will be corrected on the screen.

Then, it is determined whether or not the prediction mode PredMode of the coding block colCb is inter-prediction (MODE_INTER) (step S4020). If the prediction mode of colCb is not inter-prediction (step S4020: NO), set the flag availableFlagSbCol = 0 indicating the existence of the sub-block time merge candidate (step S4003), and end the processing of the sub-block time merge candidate derivation unit. ..

On the other hand, when the prediction mode of colCb is inter-prediction (step S4020: YES), inter-prediction information is derived for each reference list (steps S4022 and S4023). Here, for colCb, the central motion vector ctrMvLX for each reference list and the flag ctrPredFlagLX indicating whether or not LX prediction is used are derived. LX indicates a reference list, LX becomes L0 in the derivation of reference list 0, and LX becomes L1 in the derivation of reference list 1. Derivation of inter-prediction information will be described with reference to FIG. 47.

When the coding blocks colCb at different times are not available (step S4112: NO), or when the prediction mode PredMode is intra-prediction (MODE_INTRA) (step S4114: NO), both the flag availableFlagLXCol and the flag predFlagLXCol are set to 0 (step S4116). , The motion vector mvCol is set to (0,0) (step S4118), and the process of deriving the inter-prediction information is completed.

When the coding block colCb is available (step S4112: YES) and the prediction mode PredMode is not intra-prediction (MODE_INTRA) (step S4114: YES), mvCol, refIdxCol and availableFlagCol are calculated by the following procedure.

When the flag PredFlagLX [xPCol] [yPCol] indicating whether or not the LX prediction of the coded block colCb is used is 1 (step S4120: YES), the motion vector mvCol is the motion vector of the LX of the coded block colCb. The same value as MvLX [xPCol] [yPCol] is set (step S4122), the reference index refIdxCol is set to the same value as the reference index RefIdxLX [xPCol] [yPCol] of LX (step S4124), and the list listCol is set to LX. (Step S4126). Here, xPCol and yPCol are indexes indicating the positions of the upper left pixels of the coding block colCb in the picture ColPic at different times.

On the other hand, when the flag PredFlagLX [xPCol] [yPCol] indicating whether or not the LX prediction of the coding block colCb is used is 0 (step S4120: NO), the following processing is performed. First, it is determined whether or not the POCs of all the pictures registered in all the reference lists are equal to or less than the POCs of the current processing target pictures (step S4128). In addition, it is determined whether or not the flag PredFlagLY [xPCol] [yPCol] indicating whether or not the LY prediction of colCb is used is 1 (step S4128). Here, the LY prediction is defined as a reference list different from the LX prediction. That is, LY = L1 at LX = L0 and LY = L0 at LX = L1.

If this determination is true (step S4128: YES), the motion vector mvCol is set to the same value as the LY motion vector MvLY [xPCol] [yPCol] of the coding block colCb (step S4130) and the reference index refIdxCol is set. The reference index RefIdxLY [xPCol] [yPCol] of LY is set to the same value (step S4132), and the list listCol is set to LX (step S4134).

On the other hand, when this determination is false (step S4128: NO), the flag availableFlagLXCol and the flag predFlagLXCol are both set to 0 (step S4116), the motion vector mvCol is set to (0,0) (step S4118), and the inter-prediction information is derived. To finish.

When the inter-prediction information can be obtained from the coding block colCb, both the flag availableFlagLXCol and the flag predFlagLXCol are set to 1 (step S4136).

Subsequently, the motion vector mvCol is scaled to obtain the motion vector mvLXCol (step S4138). Since this process is the same as S4245 in the time prediction motion vector candidate derivation unit 322, the description thereof will be omitted.

See FIG. 44 again. After deriving the inter-prediction information for each reference list, the calculated motion vector mvLXCol is set as the central motion vector ctrMvLX, and the calculated flag predFlagLXCol is set as the flag ctrPredFlagLX (step S4022, step S4023).

Then, it is determined whether the center motion vector is valid or invalid (step S4024). If ctrPredFlagL0 = 0 and ctrPredFlagL1 = 0, it is judged to be invalid, otherwise it is judged to be invalid. When the center motion vector is invalid (step S4024: NO), the flag availableFlagSbCol = 0 indicating the existence of the subblock time merge candidate is set (step S4003), and the processing of the subblock time merge candidate derivation unit ends.

On the other hand, when the center motion vector is valid (step S4024: YES), the subblock motion information is derived by setting the flag availableFlagSbCol = 1 indicating the existence of the subblock time merge candidate (step S4025) (step S4026). This process will be described with reference to FIG. 48.

First, the number of subblocks numSbX in the width direction and the number of subblocks numSbY in the height direction are calculated from the width cbWidth and the height cBheight of the coded block colCb (step S4152). Further, refIdxLXSbCol = 0 is set (step S4152). After this processing, iterative processing is performed in units of the prediction subblock colSb. This repetition is performed while changing the index ySbIdx in the height direction from 0 to numSbY and the index xSbIdx in the width direction from 0 to numSbX.

Assuming that the upper left position of the coding block colCb is (xCb, yCb), the upper left position (xSb, ySb) of the prediction subblock colSb is
xSb = xCb + xSbIdx * sbWidth
ySb = yCb + ySbIdx * sbHeight
Is calculated. Next, the position where the temporal motion vector tempMv is added to the prediction subblock colSb is set as a new colSb (step S4154). If the upper left position of the prediction subblock colSb is (xColSb, yColSb) and the temporal motion vector tempMv is 1/16 pixel precision (tempMv [0], tempMv [1]), the upper left position of the new colSb is
xColSb = Clip3 (xCtb, xCtb + CtbSizeY + 3, xSb + (tempMv [0] >> 4))
yColSb = Clip3 (yCtb, yCtb + CtbSizeY --1, ySb + (tempMv [1] >> 4))
Will be. Here, the upper left position of the tree block is (xCtb, yCtb), and the size of the tree block is CtbSizeY. As shown in the above equation, the position after tempMv addition is corrected to a range of about the size of the tree block so that it does not deviate significantly compared to before tempMv addition. If this position is off the screen, it will be corrected on the screen.

Then, the inter-prediction information is derived for each reference list (steps S4156 and S4158). Here, for the prediction subblock colSb, the motion vector mvLXSbCol for each reference list and the flag availableFlagLXSbCol indicating whether or not the prediction subblock is valid are derived for each subblock. LX indicates a reference list, LX becomes L0 in the derivation of reference list 0, and LX becomes L1 in the derivation of reference list 1. Since the derivation of the inter-prediction information is the same as S4022 and S4023 in FIG. 47, the description thereof will be omitted.

After deriving the inter-prediction information (steps S4156, S4158), it is determined whether or not the prediction subblock colSb is valid (step S4160). If availableFlagL0SbCol = 0 and availableFlagL1SbCol = 0, colSb is invalid, otherwise it is judged to be valid. When colSb is invalid (step S4160: NO), the motion vector mvLXSbCol is set to the center motion vector ctrMvLX (step S4162). Further, the flag predFlagLXSbCol indicating whether or not the LX prediction is used is set to the flag ctrPredFlagLX in the central motion vector (step S4162). This completes the derivation of subblock motion information.

See FIG. 44 again. Then, the motion vector mvL0SbCol of L0 and the motion vector mvL1SbCol of L1 are added as candidates to the subblock merge candidate list subblockMergeCandList in the above-mentioned subblock merge mode derivation unit 304 (step S4028). However, this addition is only when the flag availableSbCol = 1, which indicates the existence of subblock time merge candidates, is set. As described above, the processing of the time merge candidate derivation unit 342 is completed.

The above description of the sub-block time merge candidate derivation unit 381 is for encoding, but the same applies for decoding. That is, the operation of the subblock time merge candidate derivation unit 481 in the subblock merge mode derivation unit 404 of FIG. 22 is similarly described by replacing the coding in the above description with decoding.

<Motion compensation prediction processing>
The motion compensation prediction unit 306 acquires the position and size of the block currently subject to prediction processing in coding. Further, the motion compensation prediction unit 306 acquires the inter prediction information from the inter prediction mode determination unit 305. The reference index and motion vector are derived from the acquired inter-prediction information, and the reference picture specified by the reference index in the decoded image memory is moved from the same position as the image signal of the prediction block by the motion vector. After acquiring the signal, the prediction signal is generated.

When the inter-prediction mode in inter-prediction is prediction from a single reference picture such as L0 prediction or L1 prediction, the prediction signal acquired from one reference picture is used as the motion compensation prediction signal, and the inter-prediction mode is BI. When the prediction mode is prediction from two reference pictures such as prediction, the weighted average of the prediction signals acquired from the two reference pictures is used as the motion compensation prediction signal, and the motion compensation prediction signal is used to determine the prediction method. Supply to the department. Here, the ratio of the weighted average of the biprediction is 1: 1, but the weighted average may be performed using another ratio. For example, the weighting ratio may be increased as the distance between the picture to be predicted and the reference picture is closer. Further, the weighting ratio may be calculated by using the correspondence table between the combination of picture intervals and the weighting ratio.

The motion compensation prediction unit 406 has the same function as the motion compensation prediction unit 306 on the coding side. The motion compensation prediction unit 406 transfers the inter-prediction information from the normal prediction motion vector mode derivation unit 401, the normal merge mode derivation unit 402, the subblock prediction motion vector mode derivation unit 403, and the subblock merge mode derivation unit 404 to the switch 408. Get through.

The motion compensation prediction unit 406 supplies the obtained motion compensation prediction signal to the decoded image signal superimposition unit 207.

<About inter-prediction mode>
The process of making a prediction from a single reference picture is defined as a simple prediction, and in the case of a single prediction, either one of the two reference pictures registered in the reference lists L0 and L1, which is L0 prediction or L1 prediction, is used. Make a prediction. The L0 prediction and the L1 prediction may be forward prediction (prediction that refers to a front reference image) or backward prediction (prediction that refers to a rear reference image). FIGS. 57 to 58 are diagrams for explaining motion compensation prediction in L0 prediction (single prediction).

FIG. 57 shows a case where the inter-prediction mode is L0 prediction and the reference picture (RefL0Pic) of L0 is at a time before the processing target picture (CurPic). FIG. 58 shows a case where the reference picture of L0 is at a time after the picture to be processed in the L0 prediction. Similarly, the reference picture for L0 prediction in FIGS. 57 and 58 can be replaced with the reference picture for L1 prediction (RefL1Pic) to perform simple prediction.

The process of making a prediction from two reference pictures is defined as a double prediction, and in the case of a double prediction, both the L0 prediction and the L1 prediction are used to express the double prediction. FIGS. 59 to 61 are diagrams for explaining motion compensation prediction in bi-prediction. FIG. 59 shows a case where the reference picture of the L0 prediction is at a time before the processing target picture and the reference picture of the L1 prediction is at a time after the processing target picture in the bi-prediction. FIG. 60 shows a case where the reference picture of the L0 prediction and the reference picture of the L1 prediction are at a time before the picture to be processed in the bi-prediction. FIG. 61 shows a case where the reference picture of the L0 prediction and the reference picture of the L1 prediction are at a time after the processing target picture in the bi-prediction.

In this way, the relationship between the prediction type of L0 / L1 and time is limited to L0 for forward prediction (prediction that refers to the front reference image) and L1 for backward prediction (prediction that refers to the rear reference image). It can be used without it. Further, in the case of bi-prediction, L0 prediction and L1 prediction may be performed using the same reference picture. It should be noted that the determination of whether the motion compensation prediction is performed by simple prediction or double prediction is determined based on, for example, information (for example, a flag) indicating whether or not to use L0 prediction and whether or not to use L1 prediction. To.

<About reference index>
In the embodiment of the present invention, in order to improve the accuracy of motion compensation prediction, it is possible to select the optimum reference picture from a plurality of reference pictures in motion compensation prediction. Therefore, the reference picture used in the motion compensation prediction is used as the reference index, and the reference index is encoded together with the coding vector in the coding stream.

<Motion compensation processing based on normal predicted motion vector mode>
As shown in the inter-prediction unit 102 on the coding side of FIG. 16, the motion compensation prediction unit 306 is used when the inter-prediction information by the normal prediction motion vector mode derivation unit 301 is selected in the inter-prediction mode determination unit 305. Acquires this inter-prediction information from the inter-prediction mode determination unit 305, derives the inter-prediction mode, reference index, and motion vector of the block currently being processed, and generates a motion compensation prediction signal. The generated motion compensation prediction signal is supplied to the prediction method determination unit 105.

Similarly, the motion compensation prediction unit 406 is normally used when the switch 408 is connected to the normal prediction motion vector mode derivation unit 401 in the process of decoding, as shown by the inter prediction unit 203 on the decoding side of FIG. The inter-prediction information by the prediction motion vector mode derivation unit 401 is acquired, the inter-prediction mode, the reference index, and the motion vector of the block currently being processed are derived, and the motion compensation prediction signal is generated. The generated motion compensation prediction signal is supplied to the decoded image signal superimposing unit 207.

<Motion compensation processing based on normal merge mode>
As shown in the inter-prediction unit 102 on the coding side of FIG. 16, the motion compensation prediction unit 306 is used when the inter-prediction information by the normal merge mode derivation unit 302 is selected in the inter-prediction mode determination unit 305. This inter-prediction information is acquired from the inter-prediction mode determination unit 305, the inter-prediction mode, reference index, and motion vector of the block currently being processed are derived, and a motion compensation prediction signal is generated. The generated motion compensation prediction signal is supplied to the prediction method determination unit 105.

Similarly, the motion compensation prediction unit 406 is in the normal merge mode when the switch 408 is connected to the normal merge mode derivation unit 402 in the decoding process, as shown by the inter prediction unit 203 on the decoding side in FIG. The inter-prediction information by the derivation unit 402 is acquired, the inter-prediction mode, the reference index, and the motion vector of the block currently being processed are derived, and the motion compensation prediction signal is generated. The generated motion compensation prediction signal is supplied to the decoded image signal superimposing unit 207.

<Motion compensation processing based on subblock prediction motion vector mode>
In the motion compensation prediction unit 306, as shown by the inter prediction unit 102 on the coding side of FIG. 16, when the inter prediction information by the subblock prediction motion vector mode derivation unit 303 is selected in the inter prediction mode determination unit 305. The inter-prediction information is acquired from the inter-prediction mode determination unit 305, the inter-prediction mode, the reference index, and the motion vector of the block currently being processed are derived, and the motion compensation prediction signal is generated. The generated motion compensation prediction signal is supplied to the prediction method determination unit 105.

Similarly, the motion compensation prediction unit 406, as shown by the inter-prediction unit 203 on the decoding side of FIG. 22, when the switch 408 is connected to the subblock prediction motion vector mode derivation unit 403 in the decoding process, The sub-block prediction motion vector mode derivation unit 403 acquires the inter-prediction information, derives the inter-prediction mode, the reference index, and the motion vector of the block currently being processed, and generates a motion compensation prediction signal. The generated motion compensation prediction signal is supplied to the decoded image signal superimposing unit 207.

<Motion compensation processing based on subblock merge mode>
As shown in the inter-prediction unit 102 on the coding side of FIG. 16, the motion compensation prediction unit 306 is used when the inter-prediction information by the sub-block merge mode derivation unit 304 is selected in the inter-prediction mode determination unit 305. , This inter-prediction information is acquired from the inter-prediction mode determination unit 305, the inter-prediction mode, the reference index, and the motion vector of the block currently being processed are derived, and the motion compensation prediction signal is generated. The generated motion compensation prediction signal is supplied to the prediction method determination unit 105.

Similarly, the motion compensation prediction unit 406 is a subblock when the switch 408 is connected to the subblock merge mode derivation unit 404 in the decoding process, as shown by the inter prediction unit 203 on the decoding side of FIG. The inter-prediction information by the merge mode derivation unit 404 is acquired, the inter-prediction mode, the reference index, and the motion vector of the block currently being processed are derived, and the motion compensation prediction signal is generated. The generated motion compensation prediction signal is supplied to the decoded image signal superimposing unit 207.

<Motion compensation processing based on affine mode>
In this embodiment, motion compensation by an affine model can be used. Motion compensation by the affine model uses 2 to 4 corners of the coded block as control points, derives the motion vector of the subblock from the motion vector of the control point, and performs motion compensation in subblock units.

The following flags are reflected in the following flags based on the inter-prediction conditions determined by the inter-prediction mode determination unit 305 in the coding process, and are encoded in the coding stream. In the decoding process, it is specified whether or not motion compensation by the affine model is performed based on the following flags in the coded stream.

sps_affine_enabled_flag indicates whether or not motion compensation by the affine model can be used in inter-prediction. If sps_affine_enabled_flag is 0, motion compensation by the affine model is suppressed in sequence units. Also, inter_affine_flag and cu_affine_type_flag are not transmitted in the CU (encoding unit) syntax of the encoded video sequence. If sps_affine_enabled_flag is 1, motion compensation by the affine model can be used in the coded video sequence.

sps_affine_type_flag indicates whether or not motion compensation by 6-parameter affine mode can be used in inter-prediction. The 6-parameter affine model is a mode in which the motion vector of a subblock is derived from the 6 parameters of the horizontal and vertical components of the motion vector of each of the three control points, and the motion is compensated for each subblock. The motion vector is derived for each sub-block, but the common reference index is derived for each coded block.

If sps_affine_type_flag is 0, it is suppressed so that it is not motion compensation by the 6-parameter affine model. Also, cu_affine_type_flag is not transmitted in the CU syntax of the encoded video sequence. If sps_affine_type_flag is 1, motion compensation with a 6-parameter affine model can be used in the encoded video sequence.

If sps_affine_type_flag does not exist, it shall be 0.

When decoding P or B slices, if inter_affine_flag is 1 in the CU currently being processed, an affine model is used to generate a motion compensation prediction signal for the CU currently being processed. Motion compensation is used.

If inter_affine_flag is 0, the affine model is not used for the CU currently being processed.

If inter_affine_flag does not exist, it shall be 0.

When decoding P or B slices, if cu_affine_type_flag is 1 in the CU currently being processed, a 6-parameter affine is used to generate a motion compensation prediction signal for the CU currently being processed. Model motion compensation is used.

If cu_affine_type_flag is 0, motion compensation by the 4-parameter affine model is used to generate the motion compensation prediction signal of the CU currently being processed. The 4-parameter affine model is a mode in which the motion vector of a subblock is derived from the four parameters of the horizontal component and the vertical component of the motion vector of each of the two control points, and the motion compensation is performed for each subblock.

<Merge differential motion vector (MMVD)>
The difference motion vector can be added to the motion vectors of the top two merge candidates (merge candidates whose merge indexes are 0 and 1 in the merge candidate list). This differential motion vector is called a merge differential motion vector.

When the merge difference motion vector is added in the merge candidate selection unit 347 on the coding side, the motion vector to which the merge difference motion vector is added is supplied to the motion compensation prediction unit 306 via the inter prediction mode determination unit 305. Further, the bit string coding unit 108 encodes the information regarding the merge difference motion vector. The information about the merged difference motion vector is an index mmvd_distance_idx indicating the distance to be added to the motion vector and an index mmvd_direction_idx indicating the direction in which the motion vector is added. These indexes are defined as shown in the tables shown in FIGS. 63 (a) and 63 (b). Then, when the x and y components of the merge difference motion vector offset MmvdOffset are expressed by MmvdOffset [0] and MmvdOffset [1], respectively,
MmvdOffset [0] = (MmvdDistance << 2) * MmvdSign [0]
MmvdOffset [1] = (MmvdDistance << 2) * MmvdSign [1]
Will be. The merged difference motion vector is derived from the merged difference motion vector offset MmvdOffset in the above equation. The details of deriving the merged difference motion vector will be described in the case of the decoding side below.

When the merge differential motion vector exists on the decoding side, the information about the merge differential motion vector is separated from the bit stream supplied to the bit string decoding unit 201, and the merge differential motion vector offset MmvdOffset is derived. Further, the merge candidate selection unit 447 derives the merge difference motion vector from the decoded merge difference motion vector offset. After adding this merge difference motion vector to the motion vector, the motion vector is supplied to the motion compensation prediction unit 406.

The derivation of the merge difference motion vector mMvdLX in the merge candidate selection unit 447 will be described with reference to the flowchart of FIG. 64 (a). First, it is determined whether or not the inter-prediction mode of the coded block is bi-prediction (PRED_BI) (S4402). If it is not a bi-prediction (S4402: No), it is determined whether or not it is an L0 prediction (PRED_L0) (S4404). In the case of L0 prediction (S4404: Yes),
mMvdL0 = MmvdOffset
mMvdL1 = 0
(S4406), the process of deriving the merge differential motion vector ends. In the case of L1 prediction (S4404: No),
mMvdL0 = 0
mMvdL1 = MmvdOffset
(S4408), the process of deriving the merge difference motion vector is completed.

On the other hand, in the case of bi-prediction (S4402: Yes), the difference between the POC of the picture to be processed and the POC of the reference picture is calculated for each reference list and is set to currPocDiffL0 and currPocDiffL1, respectively (S4410). Here, the difference between the POCs of picA and picB, DiffPicOrderCnt (picA, picB), is
DiffPicOrderCnt (picA, picB) = [picA POC]-[picB POC]
Is shown. Further, the reference picture RefPicList0 [refIdxL0] is a picture indicated by the reference index refIdxL0 of the reference list L0. Similarly, the reference picture RefPicList1 [refIdxL1] is a picture indicated by the reference index refIdxL1 of the reference list L1.

Next, it is determined whether or not -currPocDiffL0 * currPocDiffL1> = 0 (step S4412). If this determination is true (step S4412: Yes),
mMvdL0 = MmvdOffset
mMvdL1 = -MmvdOffset
(Step S4414), the process of deriving the merge differential motion vector ends. On the other hand, when this determination is false (step S4412: No),
mMvdL0 = MmvdOffset
mMvdL1 = MmvdOffset
(Step S4416). Next, it is determined whether or not the absolute value of the difference in POC from the reference list L0 is equal to or greater than the absolute value of the difference in POC from the reference list L1 (step S4418). If this determination is true (step S4418: Yes), X = 0, Y = 1 (step S4420), and the merge differential motion vector mMvdL1 of L1 is scaled (step S4424). Here, mMvdLY indicates that it is mMvdL0 when Y = 0 and mMvdL1 when Y = 1. On the other hand, when this determination is false (step S4418: No), X = 1, Y = 0 (step S4422), and the merge differential motion vector mMvdL0 of L0 is scaled (step S4424). The scaling of the merged difference motion vector mMvdLY is as shown in Fig. 64 (b).
td = Clip3 (-128, 127, currPocDiffLX)
tb = Clip3 (-128, 127, currPocDiffLY)
tx = (16384 + Abs (td) >> 1) / td
distScaleFactor = Clip3 (-4096, 4095, (tb * tx + 32) >> 6)
mMvdLY = Clip3 (-32768, 32767, Sign (distScaleFactor * mMvdLY)
* ((Abs (distScaleFactor * mMvdLY) + 127) >> 8))
Derived as. Here, it is shown that currPocDiffLX is currPocDiffL0 when X = 0 and currPocDiffL1 when X = 1. Similarly, currPocDiffLY indicates that it is currPocDiffL0 when Y = 0 and currPocDiffL1 when Y = 1. Clip3 (x, y, z) is a function that limits the minimum value to x and the maximum value to y for the value z. Sign (x) is a function that returns the sign of the value x, and Abs (x) is a function that returns the absolute value of the value x. This completes the process of deriving the merged difference motion vector.

The merge difference motion vector may be added to the top two motion vectors of the subblock merge candidates. In this case, the index mmvd_distance_idx indicating the distance to be added to the motion vector is defined as shown in the table shown in FIG. 63 (c). Since the operation of the subblock merge candidate selection unit 386 is the same as that of the merge candidate selection unit 347, the description thereof will be omitted. Further, since the operation of the subblock merge candidate selection unit 486 is the same as that of the merge candidate selection unit 447, the description thereof will be omitted.

As described above, MmvdDistance is defined as shown in the tables shown in FIGS. 63 (a) and 63 (c). Since these tables are defined with 1/4 pixel precision, the merged difference motion vector generated may include decimal pixel precision. However, by encoding / decoding the flag indicating that the pixel precision of these tables is 1 in slice units, the generated merge difference motion vector can be changed so as not to include the decimal pixel precision. ..

<Adaptive motion vector resolution (AMVR)>
The resolution of the differential motion vector can be adaptively changed for each coded block. This resolution is called the adaptive motion vector resolution.

A case where the adaptive motion vector resolution is used for the normal predicted motion vector mode will be described. In this case, in the spatial prediction motion vector candidate derivation units 321 and 421, the time prediction motion vector candidate derivation units 322 and 422, and the history prediction motion vector candidate derivation units 323 and 423, the motion vectors of the derived candidates depend on the resolution. It is rolled up. The resolution can be selected from 1/4, 1 and 4 pixel precision, and if the resolution is not changed, it will be 1/4 pixel precision. The rounding process is performed according to the resolution of the motion vector in the coded block to be processed. That is, the derived candidate motion vector mvX is
rightShift = leftShift = MvShift + 2
offset = 1 << (rightShift --1)
mvX = (mvX> = 0? (MvX + offset) >> rightShift:
-((--mvX + offset) >> rightShift)) << leftShift
Is rounded. Here, when the resolution of the motion vector in the coded block to be processed is 1/4 pixel accuracy, MvShift = 0. Similarly, when the resolution of the motion vector is 1 pixel accuracy, MvShift = 2, and when the resolution of the motion vector is 4 pixels accuracy, MvShift = 4. According to the above equation, each of the x and y components of mvX is processed.

Adaptive motion vector resolution can also be used for subblock predictive motion vector modes. In this case, only the resolution is different from the above-mentioned normal predicted motion vector mode. That is, in the affine inheritance prediction motion vector candidate derivation unit 361 and 461, the affine construction prediction motion vector candidate derivation unit 362 and 462, and the affine same prediction motion vector candidate derivation unit 363 and 463, the motion vector of the derived candidate has a resolution. It is rolled according to. The resolution can be selected from 1/16, 1/4, and 1 pixel precision, and if the resolution is not changed, the resolution will be 1/16 pixel precision. The rounding process is performed according to the resolution of the motion vector in the coded block to be processed. That is, the derived candidate motion vector mvX is rounded by the above equation. Here, when the resolution of the motion vector in the coded block to be processed is 1/4 pixel accuracy, MvShift = 0. Similarly, when the resolution of the motion vector is 1 pixel precision, MvShift = 2. According to the above formula, each of the x and y components of mvX is processed.

<Triangular merge mode of the present invention>
As one of the merge modes, there is a triangular merge mode. The triangular merge mode is a prediction mode that can be used when the use is permitted in sequence units, the B slice (or tile) is used, and the area of the coded block is 64 pixels or more.

The prediction of the triangular merge mode will be described with reference to FIG. 65. FIG. 65 shows the prediction of the coded block in the 16x16 triangular merge mode. The coded blocks in triangular merge mode are divided into 4x4 subblocks, each subblock having three single-predictive partitions 0 (UNI0), single-predictive partition 1 (UNI1), or double-predictive partition 2 (BI). Assigned to a partition. Here, the subblock on the upper side of the diagonal line is assigned to partition 0, the subblock on the lower side of the diagonal line is assigned to partition 1, and the subblock on the diagonal line is assigned to partition 2. If merge_triangle_split_dir is 0, a partition is assigned as shown in FIG. 65 (a), and if merge_triangle_split_dir is 1, a partition is assigned as shown in FIG. 65 (b).

The motion information of partition 0 is simple prediction motion information specified by the merge triangular index 0. The motion information of partition 1 is simple prediction motion information specified by the merge triangular index 1. The motion information of the partition 2 is bi-predicted motion information in which the single-predicted motion information specified by the merge triangular index 0 and the single-predicted motion information specified by the merge triangular index 1 are combined.

Here, the single-prediction motion information is a set of a motion vector and a reference index, and the bi-prediction motion information is a set of a motion vector and a reference index. Further, the motion information is simple-predicted motion information or bi-predicted motion information.

The merge candidate selection units 347 and 447 derive the triangular merge candidate list triangleMergeCandList from the derived merge candidate list mergeCandList. This process will be described with reference to FIG.

First, a part or all candidates of the merge candidate list mergeCandList are used as the triangular merge candidate list triangleMergeCandList (step S4501). Then, let the number of triangular merge candidate lists be numTriangleMergeCand.

Then, the following processing is repeated from m = 0 to numTriangleMergeCand-1. Let M be the m-th candidate in the triangleMergeCandList (step S4502). Next, it is determined whether or not the prediction mode of the candidate M is PRED_BI (step S4503). In the case of PRED_BI (step S4503: Yes), m is incremented by +1 to move to the next process. On the other hand, if it is not PRED_BI (step S4503: No), it is determined whether or not the prediction mode of the candidate M is PRED_L1 (step S4504).

In the case of PRED_L1 (step S4504: Yes), the reference index refIdxL0M = 0 and the motion vector mvL0M = (0,0) are set for the reference list L0 of the candidate M (step S4505). On the other hand, if it is not PRED_L1 (step S4504: No), the reference index refIdxL1M = 0 and the motion vector mvL1M = (0,0) are set for the reference list L1 of the candidate M (step S4506).

Spatial merge candidate derivation unit (341, 441), time merge candidate derivation unit (342, 442), history merge candidate derivation unit (345,445), average merge candidate derivation unit (344, 444), and merge candidate replenishment unit (344, 444) In the merge candidate list constructed in 346, 446), motion information candidates of single prediction and double prediction are mixed. Therefore, in the triangular merge mode, the merge candidate list is converted into the triangular merge candidate list by converting the simple-predicted motion information included in the merge candidate list into the bi-predicted motion information. The conversion from simple-predicted motion information to bi-predicted motion information assigns a predetermined reference index and a predetermined motion vector to the invalid predicted motion information.

When the screen without pan scan is fixed, the background screen is still and the objects in front often move. That is, by setting one motion vector to (0,0), the prediction efficiency can be improved when one partition is stationary and the other partition is moved. Further, in general, a reference picture having a small reference index has the highest prediction efficiency. Therefore, the prediction efficiency can be improved by setting the reference index to 0.

Here, for ease of processing, the reference indexes and motion vectors of the invalid predictions are set to 0 and (0,0), respectively. However, if the reference index and motion vector of a valid prediction are 0 and (0,0), respectively, the result is the same as a normal merge. So, to get a different effect than normal merging, the reference indexes and motion vectors of the invalid predictions are 0 and (0,0), respectively, so that the reference indexes and motion vectors of partition 0 and partition 1 are not the same. It does not have to be. For example, the reference indexes and motion vectors of invalid predictions may be 1 and (0,0), 0 and (1,1), 1 and (1,1), and so on.

Here, the reference index is fixed at 0, but the reference index is incremented by 1 according to the number of occurrences of the reference index of the prediction that is invalid for each prediction, such as 0, 1, 2, ... It is also possible to increase the selection candidates to improve the coding efficiency.

As an example of using a part of the merge candidate list, additional candidates to be replenished by the spatial merge candidate and the merge candidate replenishment unit 346 and the merge candidate replenishment unit 446, the spatial merge candidate and the time merge candidate and the additional candidate, and the history merge candidate. And there are additional candidates.

The merge candidate selection unit 447 on the decoding side acquires motion information from the derived triangular merge candidate list triangleMergeCandList. This process will be described with reference to FIG. 67.

First, a triangular merge candidate list triangleMergeCandList is generated as shown in FIG. 66 (step S4520).

Hereinafter, steps S4521 to S4522 are repeated for X = 0 to 1. From the triangular merge candidate list triangleMergeCandList, candidate M of partition X is selected based on the merge triangular index merge_triangle_idxX (step S4521).

Next, the motion vector mvLXM of the LX prediction of the candidate M and the reference index refIdxLXM are set to the motion vector mvX of the partition X and the reference index refIdxX (step S4522).

As described above, by setting all the motion information included in the triangular merge candidate list as bi-prediction, the maximum number of merge candidates and the maximum number of triangular merge candidates can be fixed to be the same. Further, the simple prediction motion information indicated by the merge triangular index 0 can be referred to as L0 prediction, and the simple prediction motion information indicated by the merge triangular index 1 can be referred to as L1 prediction. Further, the merge triangular index 0 can be associated with merge_triangle_idx0, and the merge triangular index 1 can be associated with merge_triangle_idx1 to ensure robustness at the time of decoding.

The merge candidate selection unit 347 on the coding side acquires motion information from the derived triangular merge candidate list triangleMergeCandList, and calculates the code amount and the strain amount. This process will be described with reference to FIG. 68.

First, a triangular merge candidate list triangleMergeCandList is generated as shown in FIG. 66 (step S4509).

After that, the process is repeated from X = 0 to 1, m = 0 to numTriangleMergeCand-1. First, let M be the m-th candidate of the triangular merge candidate list triangleMergeCandList (step S4510). Further, the reference index refIdxLA of partition 0 is set as the reference index refIdxLXM of the candidate M (step S4510). Further, the motion vector mvLA of partition 0 is set as the motion vector mvLXM of the candidate M (step S4510). Here, refIdxLXM is a reference index of the reference list LX, and when X = 0, it represents the reference index refIdxL0M of the reference list L0, and when X = 1, it represents the reference index refIdxL1M of the reference list L1. Further, mvLXM is a motion vector of the reference list LX, and when X = 0, it represents the motion vector mvL0M of the reference list L0, and when X = 1, it represents the motion vector mvL1M of the reference list L1.

After that, the process is repeated from n = 0 to numTriangleMergeCand-1. First, it is determined whether or not n = m (step S4511). When n = m is not satisfied (step S4511: No), the nth candidate of the triangleMergeCandList is set to N (step S4512). Further, the reference index refIdxLB of the partition 1 is set as the reference index refIdxLYN of the candidate N (step S4512). Further, the motion vector mvLB of the partition 1 is set as the motion vector mvLYN of the candidate N (step S4512). Here, Y represents 1 when X = 0 and 0 when X = 1. In addition, refIdxLYN is the reference index of the reference list LY, and when Y = 0, it represents the reference index refIdxL0N of the reference list L0, and when Y = 1, it represents the reference index refIdxL1N of the reference list L1. Further, mvLYN is a motion vector of the reference list LY, and when Y = 0, it represents the motion vector mvL0N of the reference list L0, and when Y = 1, it represents the motion vector mvL1N of the reference list L1.

Next, the code amount and the strain amount are calculated using the acquired motion information (step S4513). On the other hand, when n = m (step S4511: Yes), the processes of steps S4512 and S4513 are not performed.

By comparing a plurality of calculated code amounts and strain amounts, the direction in which the coded block is divided and the triangular merge candidate for each divided partition are selected. When encoding using the triangular merge mode, the selected information (the direction merge_triangle_split_dir that divides the coded block and the merge triangular index merge_triangle_idx0, merge_triangle_idx1 that indicates the triangular merge candidates for each divided partition), and the triangular merge candidates Inter-prediction information is supplied to the motion compensation prediction unit 306. Further, the bit string coding unit 108 encodes the selected information.

On the other hand, in the case of the triangular merge mode, the merge candidate selection unit 447 on the decoding side has the decoded information (the direction merge_triangle_split_dir for dividing the coded block and the merge triangular index indicating the triangular merge candidates for each divided partition). Based on merge_triangle_idx0, merge_triangle_idx1), a triangular merge candidate is selected, and the inter-prediction information of the selected triangular merge candidate is supplied to the motion compensation prediction unit 406.

The motion compensation prediction units 306 and 406 perform the following weighted averaging in the case of the triangular merge mode. For brightness, for the width nCbW and height nCbH of the coded block,
nCbR = (nCbW> nCbH)? (NCbW / nCbH): (nCbH / nCbW)
Is calculated. Then, at the position (x, y) in the coded block, the weight wValue in the case of FIG. 65 (a) is
wValue = (nCbW> nCbH)?
(Clip3 (0, 8, x / nCbR --y + 4)):
(Clip3 (0, 8, y / nCbR-x + 4))
Will be. On the other hand, the weight wValue in the case of FIG. 65 (b) is
wValue = (nCbW> nCbH)?
(Clip3 (0, 8, nCbH --1 --x / nCbR --y + 4))::
(Clip3 (0, 8, nCbW --1 --y / nCbR --x + 4))
Will be. Furthermore, for the number of bits bitDepth,
shift1 = max (5, 17 --bitDepth)
offset1 = 1 << (shift1-1)
Is calculated. Then, the result of the weighted average, pbSamples,
pbSamples = Clip3 (0, (1 << bitDepth) --1,
(predSamplesLA * wValue +
predSamplesLB * (8 --wValue) + offset1) >> shift1)
Will be. Here, predSamplesLA is a pixel value for which motion compensation is performed using the motion vector mvLA, and predSamplesLB is a pixel value for which motion compensation is performed using mvLB.

As described above, candidates are selected in units of merged triangular indexes, that is, candidates are selected for partition 0 and partition 1, and the motion vector and reference index of each selected candidate are predicted or doubled for each partition. By using it as a prediction, the prediction efficiency can be improved.

Here, the conversion from the merge candidate list to the triangular merge candidate list was performed by the merge candidate selection unit 347 and the merge candidate selection unit 447. For this, a list conversion unit may be installed in front of the merge candidate selection unit 347 and the merge candidate selection unit 447, and the list conversion unit may perform conversion from the merge candidate list to the triangular merge candidate list. In addition, the spatial merge candidate derivation unit (341, 441), the time merge candidate derivation unit (342, 442), the history merge candidate derivation unit (345, 445), the average merge candidate derivation unit (344, 444), and the merge candidate supplementation A merge candidate list construction unit including the unit (346, 446) may be set up.

Further, here, the process of allocating the motion information of the selected candidate to each partition was executed by the merge candidate selection unit 347 and the merge candidate selection unit 447. This may be performed by the motion compensation prediction unit 306 and the motion compensation prediction unit 406. Alternatively, a motion information allocation unit may be installed in front of the motion compensation prediction unit 306 and the motion compensation prediction unit 406 to allocate motion information.

(Modification example 1)
In the first modification, the syntax of merge_mode is different from that of the first embodiment. FIG. 69 (a) or FIG. 69 (b) is a diagram showing the syntax of merge_mode of the first modification. The difference between FIGS. 69 (a) and 69 (b) is the position of merge_idx. It differs from the first embodiment in that merge_triangle_idx0 and merge_triangle_idx1 are not encoded (or decrypted), and merge_idx is encoded (or decrypted) independently of merge_triangle_flag.

Further, in the first modification, the operation of the merge candidate selection unit 447 is different from that of the first embodiment. The operation of the merge candidate selection unit 447 of the first modification will be described. FIG. 70 is a flowchart illustrating the operation of the merge candidate selection unit 447.

First, a triangular merge candidate list triangleMergeCandList is generated (step S4525). The generation of the triangular merge candidate list is the same as in FIG. 66 in the first embodiment.

Next, the candidate M of the coded block is selected from the triangular merge candidate list triangleMergeCandList based on the merge index merge_idx (step S4526).

Hereinafter, step S4527 is repeated for X = 0 to 1.

Let the motion vector mvLXM and the reference index refIdxLXM of the LX prediction of the candidate M be the motion vector mvX and the reference index refIdxX of the partition X (step S4527).

As described above, candidates are selected in units of coded blocks, and the motion vector and reference index of the selected candidates are used as single prediction or double prediction for each partition. As a result, the number of codes related to the merged triangular index can be reduced, and the coding efficiency and the throughput of the bit string can be improved.

(Modification 2)
In the second modification, the operations of the merge candidate selection unit 347 and the merge candidate selection unit 447 are different from those of the first modification. The operation of the merge candidate selection unit 347 and the merge candidate selection unit 447 of the second modification will be described with reference to FIG. 71.

First, the merge candidate list mergeCandList is acquired (step S4530).

Next, the candidate M of the coded block is selected from the merge candidate list mergeCandList based on the merge index merge_idx (step S4531).

Next, the LX prediction (X = 0,1) is repeated from step S4532 to step S4534.

Check if the LX prediction of candidate M is invalid (step S4532). If the LX prediction of the candidate M is invalid (YES in step S4532), the motion vector mvX of the LX prediction is (0,0), and the reference index refIdxX of the prediction X is set to 0 (step S4533). If the LX prediction of the candidate M is not invalid (NO in step S4532), the motion vector mvX of the LX prediction is the motion vector mvLXM of the candidate M, and the reference index refIdxX of the prediction X is the reference index refIdxLXM of the candidate M (step S4534). ..

As described above, in the case of the triangular merge mode, candidate M is selected using the merge candidate list, and only the selected candidate is converted into a candidate for bi-prediction motion information, thereby becoming a candidate for bi-prediction motion information. The conversion process can be reduced. In addition, by not generating the triangular merge candidate list triangleMergeCandList, the circuit up to the selection of candidates can be shared with the normal merge.

In step S4533 of the second modification, the motion vector of the LX prediction in which the candidate M is invalid is set to (0,0). For example, the motion vector mvX of the prediction in which the candidate M is valid is set to (0,0). It may be −mvX, which is a motion vector that is symmetrical with respect to. In -mvX, both the horizontal component and the vertical component are symmetric with respect to (0,0), but only the horizontal component is symmetric with respect to (0,0), and only the vertical component is symmetric with respect to (0,0). May be good.

Here, the conversion of the bi-prediction into motion information was performed by the merge candidate selection unit 347 and the merge candidate selection unit 447. This may be done by installing a bi-prediction conversion unit after the merge candidate selection unit 347 and the merge candidate selection unit 447, and converting the bi-prediction into motion information by the bi-prediction conversion unit.

A plurality of the above-described embodiments may be combined.

In all the embodiments described above, the coded bitstream output by the image coding apparatus has a specific data format so that it can be decoded according to the coding method used in the embodiment. are doing. The coded bit stream may be recorded and provided on a recording medium that can be read by a computer such as an HDD, SSD, flash memory, or optical disk, or may be provided from a server through a wired or wireless network. Therefore, the image decoding device corresponding to this image coding device can decode the coded bit stream of this specific data format regardless of the providing means.

When a wired or wireless network is used to exchange the coded bitstream between the image encoding device and the image decoding device, the coded bitstream is converted into a data format suitable for the transmission mode of the communication path. It may be transmitted. In that case, a transmission device that converts the coded bit stream output by the image coding device into coded data in a data format suitable for the transmission mode of the communication path and transmits the coded data to the network, and a transmission device that receives the coded data from the network and receives the coded data. A receiving device that restores the coded bitstream and supplies it to the image decoding device is provided. The transmission device includes a memory that buffers the coded bit stream output by the image coding device, a packet processing unit that packets the coded bit stream, and a transmission unit that transmits the packetized coded data via the network. And include. The receiving device generates a coded bit stream by packet processing the coded data and a receiver that receives the coded data packetized via the network, a memory that buffers the received coded data, and the coded data. It includes a packet processing unit provided to the image decoding device.

When a wired or wireless network is used to exchange a coded bitstream between the image encoding device and the image decoding device, in addition to the transmitting device and the receiving device, the coded data transmitted by the transmitting device is further transmitted. A relay device that receives and supplies to the receiving device may be provided. The relay device includes a receiving unit that receives the packetized encoded data transmitted by the transmitting device, a memory that buffers the received encoded data, and a transmitting unit that transmits the packetized encoded data to the network. Including. Further, the relay device packets the received packet processing unit that packet-processes the packetized coded data to generate the coded bit stream, the recording medium that stores the coded bit stream, and the coded bit stream. A transmission packet processing unit may be included.

Further, the display device may be used by adding a display unit for displaying the image decoded by the image decoding device to the configuration. In that case, the display unit reads the decoded image signal generated by the decoded image signal superimposing unit 207 and stored in the decoded image memory 208 and displays it on the screen.

Further, by adding an image pickup unit to the configuration and inputting the captured image to the image coding device, the image pickup device may be used. In that case, the image pickup unit inputs the captured image signal to the block division unit 101.

The above processing related to encoding and decoding may be realized as a transmission, storage, and receiving device using hardware, and is stored in a ROM (read-only memory), a flash memory, or the like. It may be realized by firmware or software such as a computer. The firmware program and software program may be recorded on a recording medium readable by a computer or the like and provided, or may be provided from a server via a wired or wireless network, or terrestrial or satellite digital broadcasting data broadcasting. May be provided as.

The present invention has been described above based on the embodiments. Embodiments are examples, and it will be understood by those skilled in the art that various modifications are possible for each of these components and combinations of each processing process, and that such modifications are also within the scope of the present invention. ..

100 image coding device, 101 block division unit, 102 inter-prediction unit, 103 intra-prediction unit, 104 decoded image memory, 105 prediction method determination unit, 106 residual signal generation unit, 107 orthogonal conversion / quantization unit, 108-bit string code Quantization unit, 109 inverse quantization / inverse orthogonal conversion unit, 110 decoded image signal superimposition unit, 111 coded information storage memory, 200 image decoding device, 201 bit string decoding unit, 202 block division unit, 203 inter prediction unit 204 intra prediction unit , 205 Coded information storage memory, 206 Inverse quantization / inverse orthogonal conversion section, 207 Decoded image signal superimposition section, 208 Decoded image memory.

Claims (1)

An image coding device that uses merge mode.
A merge candidate list construction unit that builds a merge candidate list that includes spatial merge candidates,
A merge candidate selection unit that selects one select merge candidate from the merge candidate list based on the merge index, and a merge candidate selection unit.
When the selected merge candidate is simple prediction motion information, a bi-prediction conversion unit that converts the selected merge candidate into bi-prediction motion information consisting of L0 prediction and L1 prediction,
The coded block is divided into 4x4 sub-blocks, and for each sub-block, the motion information of the L0 prediction of the twin prediction, the movement information of the L1 prediction of the twin prediction, and the movement of both the L0 prediction and the L1 prediction of the twin prediction. An image coding device having a motion information assigning unit for assigning any of the information.

Images