JP2018074182A - Encoding device, encoding method, and encoding program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a moving image encoding device that reduces a calculation amount and memory access when a prediction image is generated from an encoding object frame and can increase the encoding speed of a moving image.SOLUTION: An encoding device includes prediction image generation means 123 that generates a prediction image of an encoding target block by using a reference image and a candidate of movement information associated with the reference image from a first buffer 14 in which movement information associated with an encoded block is stored, evaluation cost calculation means 121 that calculates evaluation cost by using the encoding target block and the prediction image, a second buffer 124 that stores the associated movement information and the evaluation cost by using the movement information that is a candidate to be associated with the encoding target block as an index, and index selection means 122 that updates the index stored in the second buffer on the basis of the evaluation cost and selects the movement information that is a candidate to be associated with the encoding target block from the second buffer on the basis of the evaluation cost.SELECTED DRAWING: Figure 5

Description

  The present invention relates to an encoding device, an encoding method, and an encoding program for encoding a moving image. In particular, the present invention relates to an encoding device, an encoding method, and an encoding program that compress and encode moving image information using temporal and spatial correlation.

  H. H.264 / MPEG-4 AVC (Moving Picture Experts Group-4 Advanced Video Coding: hereinafter H.264) is one of the standards for moving picture coding. H. H.264 newer than H.264. H.265 / HEVC (High Efficiency Video Coding: hereinafter H.265) It is a standard that can reproduce the same image quality with a smaller data size than H.264. H. H.264 and H.H. In H.265, the encoding target frame constituting the moving image is divided into a plurality of pixel blocks, and encoding is performed for each block constituting the encoding target frame.

  H. In 265, information is compressed using temporal and spatial correlation. A mode in which information is compressed using temporal correlation is called inter prediction, and a mode in which information is compressed using spatial correlation is called intra prediction. For example, in the inter prediction mode, unidirectional prediction in which a prediction image is generated using one encoded reference frame and bidirectional prediction in which a prediction image is generated using two encoded reference frames are used. There are types.

  In inter prediction, motion compensated prediction is used to cope with motion between frames. Motion compensated prediction is one of the methods for compressing information using temporal correlation in inter prediction. In motion compensation prediction, an amount of motion between frames (hereinafter referred to as a motion vector) is estimated, and an image at a position shifted by the amount of motion is used as a predicted image. In motion compensated prediction, a prediction image is generated from neighboring frames for each block, and a difference between the input image and the prediction image is encoded. The motion vector is also abbreviated as MV (Motion Vector) and is encoded simultaneously with the difference image.

  H. In motion compensated prediction of H.265, a motion vector with a quarter-pixel accuracy can be used as in Non-Patent Document 1. A predicted image with decimal pixel accuracy is generated by interpolating pixels between integer pixels. FIG. 18 is a conceptual diagram showing the positions of integer pixels and decimal pixels in each pixel. In FIG. 18, the position indicated by capital letters is the position of integer pixels, and the position indicated by small letters is the position of decimal pixels. In the example of FIG. 18, interpolation is performed with a filter using a plurality of pixels located around the decimal pixel.

  H. 265 includes a merge mode as a motion vector data compression tool, as in Non-Patent Document 2. The merge mode is a method in which motion information is integrated with neighboring blocks of the encoding target block. In the merge mode, one of the motion vectors (referred to as merge candidate vectors) of the selected candidate block among the candidate block positions (hereinafter referred to as candidate blocks) for integrating the motion information is copied and used as it is. Then, only the index representing the position of the copy source candidate block is encoded.

  Patent Document 1 discloses a moving image encoding technique in which an encoding target frame is divided into blocks and each divided block is encoded using motion compensated prediction.

  In the technique of Patent Document 1, a motion vector is obtained from a plurality of encoded reference blocks at predetermined neighboring positions in the same image with respect to a prediction target block that is a prediction target of a motion vector in an encoding target frame. Extract. In the technique of Patent Document 1, a predetermined encoded image is used as a reference image, and among the blocks in the reference image having the same arrangement relationship as the reference block, the block having the smallest motion vector divergence is used. Find the area. Furthermore, in the technique of Patent Document 1, a motion vector assigned to the obtained area is extracted as a prediction vector candidate for the block.

  Patent Document 2 discloses a moving image encoding apparatus using a merge mode. The apparatus of Patent Literature 2 selects merge candidate vectors of five candidate blocks located around the encoding target block in a predetermined order. The apparatus of Patent Literature 2 sequentially determines whether or not to register the motion vector of each candidate block as a merge candidate vector. According to the apparatus of Patent Literature 2, the code amount of motion vector data can be reduced by using the merge mode.

  H. In the H.264 reference encoder JM (Joint Model), every time a motion vector evaluation cost is calculated, an interpolated image is generated on the entire screen before starting the motion vector determination for each frame without interpolating decimal pixels and creating a predicted image. And save it in memory.

  In JM, one interpolation image is created for each decimal pixel position. The predicted image is created by copying an image from an interpolation image created in advance. That is, in JM, including integer pixels, it is necessary to create a total of 16 interpolated images for each of the decimal pixels from A to r in FIG. 18, and to create predicted images from these interpolated images.

JP 2012-129756 A JP2015-173404A

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, "Overview of the High Efficiency Video Coding (HEVC) Standard" VOL. 22, NO.12, DECEMBER 2012, p.1649-1668 Information Source Coding Department, “H.265 | MPEG-H HEVC Outline of Standard Ver.1.0”, Radio Industry, Digital Broadcasting System Development Group, Video Coding System Working Group, May 2013, p . 19-20

  As in Non-Patent Document 1, when the merge candidate vector has decimal precision, decimal pixel interpolation processing is involved in creating a predicted image. As shown in FIG. 18, in order to interpolate decimal pixels, it is necessary to refer to integer pixels outside the processing block and interpolation results outside the processing block, so it is necessary to access and calculate the memory including the pixels outside the block. .

  H. In general merge selection in the 256 reference encoder HM, prediction images of merge candidate vectors for each encoding target block are generated independently. In general merge selection, the same pixel is accessed redundantly and the same fractional pixel is generated redundantly when generating a prediction image of a neighboring encoding target block.

  H. Even in a method in which a decimal interpolation image is created in advance as in the H.264 reference encoder JM, memory access efficiency is insufficient in addition to memory access duplication when a merge candidate vector is selected. Memory access efficiency can be improved by accessing consecutive memory areas at once. However, in general merge selection, it is necessary to access memory independently even for predicted images of consecutive decimal interpolation images. .

  In order to solve the above-described problems, an object of the present invention is to provide an encoding device capable of reducing the amount of calculation and memory access when generating a predicted image from an encoding target frame and speeding up the encoding of a moving image. There is to do.

  In one aspect of the present invention, an encoding device receives, as input, an encoding target block included in an encoding target frame and a reference image corresponding to the encoding target block, and motion information associated with the encoded block A prediction image generating means for acquiring a motion information candidate associated with a reference image from a first buffer in which is stored, and generating a prediction image using the motion information candidate and the reference image; a coding target block; Using the predicted image as an input, evaluation cost calculating means for calculating the evaluation cost of the coding target block using the coding target block and the predicted image, and motion information as a candidate to be associated with the coding target block as an index A second buffer for storing the information and the evaluation cost in association with each other, and a second buffer based on the evaluation cost calculated for the motion information. It has been updated the index includes an index selector for selecting from the second buffer on the basis of motion information which are candidates to be associated with the encoding target block in the evaluation cost.

  In one aspect of the present invention, an encoding method receives motion information associated with an encoded block by inputting an encoding target block included in the encoding target frame and a reference image corresponding to the encoding target block. Motion information associated with the reference image is acquired from the first buffer in which is stored, a prediction image is generated using the reference image and the motion information, an evaluation cost is calculated for the encoding target block, and the encoding target The motion information that is a candidate to be associated with the block is used as an index, the motion information and the evaluation cost are associated with each other and stored in the second buffer, and the index stored in the second buffer based on the evaluation cost calculated for the motion information is stored. The motion information that becomes a candidate to be updated and associated with the encoding target block is selected from the second buffer based on the evaluation cost.

  In one aspect of the present invention, an encoding program is associated with a process of inputting an encoding target block included in an encoding target frame and a reference image corresponding to the encoding target block, and the encoded block. A process for obtaining motion information associated with a reference image from a first buffer in which motion information is stored, a process for generating a predicted image using the reference image and the motion information, and an evaluation cost for the encoding target block Based on the processing to calculate, the processing to store the motion information and the evaluation cost in association with the motion information that is a candidate to be associated with the block to be encoded in the second buffer, and the evaluation cost calculated for the motion information Processing for updating the index stored in the second buffer, and motion information as a candidate to be associated with the encoding target block To execute a process of selecting from the second buffer, based on the evaluation cost computer.

  According to the present invention, it is possible to provide an encoding device capable of reducing the amount of calculation and memory access when generating a predicted image from an encoding target frame, and speeding up the encoding of a moving image.

It is a conceptual diagram which shows the structure of the moving image encoder which concerns on the 1st Embodiment of this invention. It is a conceptual diagram which shows an example of the flow of the signal in the moving image encoder which concerns on the 1st Embodiment of this invention. It is a conceptual diagram for demonstrating the merge mode which the moving image encoder which concerns on the 1st Embodiment of this invention uses. It is a conceptual diagram which shows the candidate block of the encoding object block which the moving image encoder which concerns on the 1st Embodiment of this invention processes. It is a block diagram which shows the structure of the motion vector determination means of the moving image encoder which concerns on the 1st Embodiment of this invention. It is a conceptual diagram which shows an example of the signal flow in the motion vector determination means of the moving image encoder which concerns on the 1st Embodiment of this invention. It is a conceptual diagram which shows the example of a division | segmentation of the encoding object block which is a process target of the moving image encoder which concerns on the 1st Embodiment of this invention. It is a conceptual diagram for demonstrating the movement of the reference image at the time of producing | generating a prediction image from the encoding object block which is a process target of the moving image encoder which concerns on the 1st Embodiment of this invention. It is a conceptual diagram which shows that a required pixel overlaps when producing | generating the prediction image of the some merge candidate vector which makes a motion vector common. It is a flowchart which shows operation | movement of the merge selection means contained in the motion vector determination means of the moving image encoder which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of the motion vector determination means of the moving image encoder of related technology. It is a flowchart for demonstrating operation | movement of the merge selection means contained in the motion vector determination means of the moving image encoding device of related technology. It is a conceptual diagram for demonstrating the general moving image encoding method. It is a flowchart which shows the operation | movement in the moving image encoding method which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the structural example of the hardware which implement | achieves the moving image encoder which concerns on each embodiment of this invention. It is a block diagram which shows another structural example of the hardware which implement | achieves the moving image encoder which concerns on each embodiment of this invention. It is a conceptual diagram which shows the structural example of GPU (Graphic Processing Unit) contained in the hardware which implement | achieves the moving image encoder which concerns on each embodiment of this invention. It is a conceptual diagram for demonstrating the integer pixel and decimal pixel in an encoding object frame.

  EMBODIMENT OF THE INVENTION Below, the form for implementing this invention is demonstrated using drawing. However, the preferred embodiments described below are technically preferable for carrying out the present invention, but the scope of the invention is not limited to the following. In all the drawings used for the description of the following embodiments, the same parts are denoted by the same reference numerals unless otherwise specified. In the following embodiments, repeated description of similar configurations and operations may be omitted. Moreover, the direction of the arrow in the drawing shows an example, and does not limit the direction of the signal between the components.

(First embodiment)
(Constitution)
First, a moving picture encoding apparatus (also referred to as an encoding apparatus) according to a first embodiment of the present invention will be described in detail with reference to the drawings.

  The moving picture encoding apparatus according to the present embodiment is an H.264 standard. When encoding an encoding target image (hereinafter referred to as an encoding target frame) included in a moving image, the encoding target frame is divided into a plurality of blocks (hereinafter referred to as encoding target blocks). ). In the present embodiment, the generation of spatially continuous predicted images across a plurality of encoding target blocks is collectively performed, thereby reducing processing and memory access that overlap between spatially continuous encoding target blocks.

  FIG. 1 is a conceptual diagram showing a configuration of a video encoding device 1 according to the present embodiment. FIG. 2 is a conceptual diagram showing arrows the signal flow when the moving picture encoding apparatus 1 of the present embodiment operates. Note that the signal flow in FIG. 2 is an example, and does not limit the signal flow in the video encoding device 1 of the present embodiment.

  The moving image encoding apparatus 1 includes a motion vector determination unit 10, an inter prediction unit 21, an intra prediction unit 22, an adder 23, a transform / quantization unit 24, an encoding unit 25, an inverse transform / inverse quantization unit 26, and a synthesis. Device 27, loop filter 28, and frame buffer 29. In particular, the moving image coding apparatus 1 is characterized by the motion vector determination means 10. The motion vector is also abbreviated as MV (Motion Vector).

The motion vector determination unit 10 (also referred to as motion information determination unit) acquires the encoding target frame as an input image. The motion vector determination unit 10 reads an encoded block (hereinafter referred to as a reference image) corresponding to any of the encoding target blocks constituting the acquired encoding target frame from the frame buffer 29.
The motion vector determination unit 10 determines motion information (motion vector) indicating the motion between the encoding target block and the reference image for the encoding target block. The motion vector determination unit 10 outputs information on the determined motion vector to the inter prediction unit 21.

  The motion vector determination means 10 duplicates motion vector candidates (hereinafter referred to as candidate motion vectors) from spatially continuous encoded blocks in order to estimate the motion between frames of the encoding target block. The motion vector determination unit 10 calculates an evaluation cost for the copied candidate motion vector.

  Then, the motion vector determination unit 10 selects a motion vector of the encoding target block from the candidate motion vectors based on the calculated evaluation cost. In the present embodiment, for example, the motion vector determination unit 10 selects a motion vector with the lowest evaluation cost.

  For example, the motion vector determination unit 10 calculates the sum of the image distortion cost due to encoding and the code amount cost necessary for encoding as the evaluation cost. For example, the motion vector determination unit 10 uses an absolute value difference sum SAD (Sum of Absolute Difference) value of pixels representing the similarity between the input image block and the predicted image block as the image distortion cost. In addition, the motion vector determination unit 10 uses a code amount for encoding a motion vector as a code amount cost. Since the similarity is higher as the SAD value is smaller, the motion vector determination unit 10 selects a motion vector that minimizes the evaluation cost.

  The inter prediction unit 21 generates a prediction image corresponding to the motion vector determined by the motion vector determination unit 10. The inter prediction unit 21 outputs the generated predicted image to the adder 23 and the combiner 27.

  The intra prediction unit 22 generates a prediction image based on the intra prediction. The intra prediction means 22 outputs the generated predicted image to the adder 23 and the combiner 27.

  The adder 23 receives the input image and the predicted image, and generates a residual signal that is a difference between the input image and the predicted image. The adder 23 outputs the generated residual signal to the transform / quantization means 24.

  The transform / quantization means 24 receives the residual signal from the adder 23 as an input, converts the residual signal into an integer, and generates a quantized transform coefficient. The transform / quantization unit 24 outputs the quantized transform coefficient to the encoding unit 25 and the inverse transform / inverse quantization unit.

  The encoding unit 25 encodes the transform coefficient quantized by the transform / quantization unit 24 and outputs the bit stream as an encoded stream.

  The inverse transform / inverse quantization unit 26 receives the transform coefficient quantized by the transform / quantization unit 24 and decodes the prediction residual image by performing inverse transform on the transform coefficient. Specifically, the inverse transform / inverse quantization means 26 decodes the prediction residual image by performing inverse integer transform after dequantizing the quantized transform coefficient.

  The synthesizer 27 receives the prediction residual image decoded by the inverse transform / inverse quantization means 26 and the prediction image generated by the inter prediction means 21 as inputs. The synthesizer 27 synthesizes the prediction residual image and the prediction image to generate a reconstructed image. The synthesizer 27 outputs the generated reconstructed image to the intra prediction unit 22 and the loop filter 28.

  The reconstructed image generated by the synthesizer 27 is used for intra prediction of the same frame in the intra prediction means 22. Further, the image filtered by the loop filter 28 is stored in the frame buffer 29 and used for inter prediction of the subsequent frame.

  The loop filter 28 removes block distortion from the reconstructed image generated by the synthesizer 27 and stores the reconstructed image from which block distortion has been removed in the frame buffer 29.

  The frame buffer 29 is a buffer for storing a reconstructed image (hereinafter referred to as a reference image) from which block distortion has been removed by the loop filter 28. The reference image stored in the frame buffer 29 is read out as a reference image from the motion vector determination unit 10 when performing inter prediction of the encoding target block of the subsequent frame.

[Merge mode]
The moving image encoding apparatus 1 of the present embodiment compresses moving image information using temporal and spatial correlation. A mode in which moving picture information is compressed using temporal correlation is called inter prediction, and a mode in which moving picture information is compressed using spatial correlation is called intra prediction.

  The inter prediction includes a motion search mode (first mode) and a merge mode (second mode). In the motion search mode, the motion vector is directly encoded. On the other hand, in the merge mode, the motion vector of the encoded block preceding the encoding target block is copied, and the copied motion vector is used as the motion vector of the encoding target block. That is, in the merge mode, the motion vector of the block referenced by the encoding target block is duplicated.

  FIG. 3 is a conceptual diagram for explaining the merge mode. The upper left frame in FIG. 3 is the encoding target frame 100. The upper right frame in FIG. 3 is a reference frame 200 that is referred to in order to calculate a motion vector. The reference frame 200 is an encoded frame.

  The encoding target block 101 included in the encoding target frame 100 is illustrated in an enlarged manner inside the broken line on the lower left side of FIG. In the lower right broken line in FIG. 3, a block (hereinafter referred to as a reference block 201) at the same position as the encoding target block 101 in the reference frame 200 is enlarged and illustrated. The seven blocks shown around the encoding target block 101 and the reference block 201 inside the broken lines on the lower left side and the lower right side in FIG. 3 are block position candidates (hereinafter referred to as candidate blocks) for integrating motion information. Show.

  In the merge mode, motion vectors (referred to as merge candidate vectors) of five candidate blocks out of the seven candidate blocks are selected. In merge mode, a copy of one of the motion vectors (called merge candidate vectors) of the selected candidate block is used as it is, and only the index (also called the merge index) that indicates the position of the copy source candidate block is encoded. Turn into. In the following description, as shown in FIG. 4, four candidate blocks (Cand2 to 5) located on the upper right side (2), upper side (3), left side (4), and upper left side (5) of the encoding target block 101. Focus only on. Cand is an abbreviation for Candidate.

[Motion vector determination means]
Here, a detailed configuration of the motion vector determination unit 10 will be described with reference to FIG. FIG. 5 is a block diagram showing a detailed configuration of the motion vector determination means 10. FIG. 6 is a conceptual diagram showing the flow of signals when the motion vector determination means 10 operates by arrows. The signal flow in FIG. 6 is an example, and the signal flow in the motion vector determination unit 10 is not limited.

  The motion vector determination means 10 includes a motion vector determination function of a general HEVC (High Efficiency Video Coding) reference encoder HM (HEVC Test Model). The motion vector determination means 10 is different from a general HEVC reference encoder HM in that a buffer for storing information on merge candidate vectors is added.

  As shown in FIG. 5, the motion vector determination unit 10 includes a motion search unit 11, a merge selection unit 12, a motion vector comparison unit 13, and a motion vector buffer 14.

  The motion search means 11 (also referred to as search means) obtains an encoding target block and a reference image of the encoding target block, and determines a motion vector in the motion search mode. The motion search means 11 outputs the determined motion vector to the motion vector comparison means 13.

  The merge selection unit 12 (also referred to as selection unit) acquires the encoding target block and the reference image, and selects a motion vector in the merge mode. The merge selection unit 12 selects a motion vector of an encoded block adjacent to the encoding target block from among the merge candidate vectors stored in the motion vector buffer 14.

  The motion vector comparison means 13 (also referred to as comparison means) compares the motion vector determined in the motion search mode and the motion vector selected in the merge mode, and uses the one with the smaller evaluation cost as the motion vector of the encoding target block. select. The motion vector comparison unit 13 outputs the selected motion vector to the inter prediction unit 21 and stores it in the motion vector buffer 14.

  The motion vector buffer 14 (also referred to as the first buffer) stores the motion vector selected by the motion vector comparison means 13. The motion vector stored in the motion vector buffer 14 is used as a merge candidate vector for a block adjacent to the subsequent encoding target block. In other words, the motion vector buffer 14 stores the motion vectors of the encoded frame and the encoded block of the encoding target frame.

[Merge selection means]
Here, a detailed configuration of the merge selection unit 12 will be described. As shown in FIGS. 5 and 6, the merge selection unit 12 includes an evaluation cost calculation unit 121, a merge index selection unit 122, a merge candidate predicted image generation unit 123, and a merge candidate buffer 124. 5 and 6 show only the characteristic configuration of the merge selection means 12, and not all of the merge selection means 12.

  The evaluation cost calculation unit 121 receives a coding target block and a predicted image generated from a reference image corresponding to the coding target block. The evaluation cost calculation unit 121 calculates the evaluation cost of the encoding target block using the input image and the predicted image.

  The evaluation cost calculation unit 121 calculates evaluation costs for various candidate motion vectors in order to estimate the motion between frames of the encoding target block.

  For example, the evaluation cost calculation unit 121 calculates the sum of the image distortion cost due to encoding and the code amount cost necessary for encoding as the evaluation cost. For example, the evaluation cost calculation unit 121 uses the absolute value difference sum SAD value of pixels representing the similarity between the input image block and the predicted image block as the image distortion cost. For example, the evaluation cost calculation unit 121 uses a code amount for encoding a motion vector as a code amount cost.

  The merge index selection unit 122 (also referred to as index selection unit) updates the best merge index stored in the merge candidate buffer 124 based on the evaluation cost calculated for the motion vector. The merge index is an index for identifying a merge candidate vector stored in the merge candidate buffer 124.

  When the best merge index is updated, the merge index selection unit 122 selects a motion vector associated with the encoding target block from the merge candidate buffer 124 based on the evaluation cost. The merge index selection unit 122 outputs the selected motion vector to the motion vector comparison unit 13.

  If the evaluation cost of the block is not stored in the merge candidate buffer 124, the merge index selection unit 122 stores the evaluation cost of the block to be encoded as the best merge index in association with the merge candidate vector in the merge candidate buffer 124. .

  On the other hand, when the evaluation cost of the block is already stored in the merge candidate buffer 124, the merge index selection unit 122 determines the newly calculated evaluation cost and the evaluation stored in the merge candidate buffer 124 for the block. Compare costs. Then, the merge index selection unit 122 updates the best merge index stored in the merge candidate buffer 124 when the evaluation cost newly calculated for the block is minimum.

  The merge candidate predicted image generation unit 123 (also referred to as a predicted image generation unit) reads a reference image corresponding to the encoding target block from the frame buffer 29. Then, the merge candidate predicted image generation unit 123 acquires a motion vector (merge candidate vector) corresponding to the input encoding target block from the motion vector buffer 14. The merge candidate predicted image generation unit 123 generates a predicted image using the reference image and the merge candidate vector. When the merge candidate vector is a decimal precision vector, the merge candidate prediction image generation unit 123 generates a prediction image by interpolating decimal pixels.

  The merge candidate predicted image generation unit 123 collectively generates predicted images corresponding to blocks to be encoded with a common motion vector as a merge candidate vector. That is, the merge candidate predicted image generation unit 123 collectively generates predicted images of a plurality of encoding target blocks using the motion vector selected as the replication source as a merge candidate vector.

  The merge candidate buffer 124 (also referred to as a second buffer) is a buffer that stores a merge candidate vector of each encoding target block and its evaluation cost in association with each other. That is, the merge candidate buffer 124 stores the motion information of the encoding target block and the evaluation cost in association with each other using the merge candidate vector of the encoding target block as an index.

[General merge mode]
Here, a general process for selecting the merge mode will be described with reference to FIGS. FIG. 7 shows an encoding target frame 100 divided into 12 encoding target blocks (B1 to B12). FIG. 8 is a conceptual diagram for explaining the movement of the reference image when the predicted image 301 is generated from the encoding target block 101. FIG. 9 is a conceptual diagram showing that necessary pixels 401 overlap when generating predicted images of a plurality of merge candidate vectors having a common motion vector. In the following, processing is performed in the order of B2, B3,..., B12 with the block B1 in the upper left of FIG.

  When generating the predicted image 301 from the encoding target block 101, first, the process of the block B1 is performed. Since the block B1 is located at the upper left corner of the encoding target block 101 and there is no adjacent block on the left side or the upper side, the merge selection process is not performed.

  In the next block B2, there is an adjacent block (block B1) on the left side, and therefore processing is performed on the merge candidate vector B2_Cand4 for copying the motion vector (hereinafter referred to as B1_MV) of the block B1. The merge candidate vector B2_Cand4 (hereinafter referred to as B2_Cand4) of the block B2 is B1_MV. Thereafter, merge selection processing is performed for the blocks B3 to B12.

  Here, as an example, the merge selection processing of blocks B7, B9, B10, and B11 in FIG. 7 will be described. In the following, the merge candidate vector Cand4 (left side) of the block B7 is denoted as B7_Cand4, and the merge candidate vector Cand2 (upper right side) of the block B9 is denoted as B9_Cand2. Similarly, the merge candidate vector Cand3 (upper side) of the block B10 is expressed as B10_Cand3, and the merge candidate vector Cand5 (upper left side) of the block B11 is expressed as B11_Cand5.

  In FIG. 8, B7_Cand4, B9_Cand2, B10_Cand3, and B11_Cand5 are all copies of the motion vector B6_MV of the block 6.

  Therefore, as shown in FIG. 8, the predicted images of B7_Cand4, B9_Cand2, B10_Cand3, and B11_Cand5 are obtained by moving the reference image by B6_MV, and are continuous regions. Therefore, as shown in FIG. 9, pixels (necessary pixels 401) necessary for predictive image generation of B7_Cand4, B9_Cand2, B10_Cand3, and B11_Cand5 overlap.

  That is, in general merge selection, when the predicted images of B7_Cand4, B9_Cand2, B10_Cand3, and B11_Cand5 are generated, the same pixel is accessed redundantly, and the same fractional pixel is generated redundantly. Necessary.

  On the other hand, in the present embodiment, the same motion vector is copied, and the copied motion vector is stored in the merge candidate buffer 124. Then, prediction images of a plurality of encoding target blocks having the same merge candidate vector are generated together. Therefore, according to the present embodiment, generation of intermediate data and memory access in the decimal pixel interpolation process can be reduced.

(Operation)
Next, the merge selection operation by the merge selection means 12 included in the moving image encoding apparatus 1 of the present embodiment will be described with reference to the flowchart of FIG. In the following description, the merge selection unit 12 is mainly described with respect to the encoding target block 101 constituting the encoding target frame 100 shown in FIGS.

  In FIG. 10, the merge selection means 12 first selects B1_MV, which is the motion vector of the block B1, as the copy source MV. B1_MV is merge candidate vectors B2_Cand4, B5_Cand3, and B6_Cand5 of the blocks B2, B5, and B6.

  The merge selection means 12 uses B1_MV to collectively generate prediction images of merge candidate vectors B2_Cand4, B5_Cand3, and B6_Cand5 that share a motion vector (step S11).

  For the generation of the predicted image in step S11, see H.264. Like the reference encoder HM of H.265, decimal pixel interpolation may be performed on the spot for each motion vector. In addition, regarding the generation of the predicted image in step S11, H.264. It may be read out from an interpolation image created in advance like a H.264 reference encoder JM (Joint Model).

  Next, the merge selection unit 12 performs the processes of steps S12 to S13 for each merge candidate vector. Here, the merge selection unit 12 performs processing in the order of merge candidate vectors B2_Cand4, B5_Cand3, and B6_Cand5.

  First, the merge selection unit 12 calculates an evaluation cost for the merge candidate vector B2_Cand4 (step S12).

  Then, the merge selection unit 12 stores the calculated evaluation cost in the merge candidate buffer 208 in association with the merge candidate vector B2_Cand4 as the best merge index of the block B2 (step S13).

  Similarly, the merge selection unit 12 performs the processes of steps S12 to S13 for the merge candidate vectors B5_Cand3 and B6_Cand5. When the process of the merge candidate vector B6_Cand5 is finished, the process of setting B1_MV as the copy source MV is finished.

  Subsequently, the merge selection unit 12 selects B2_MV, which is the motion vector of the block B2, as the copy source MV. B2_MV is merge candidate vectors B3_Cand4, B5_Cand2, B6_Cand3, and B7_Cand5 of the blocks B3, B5, B6, and B7.

  Similarly to B1_MV, merge selection means 12 uses B2_MV to collectively generate prediction images of merge candidate vectors B3_Cand4, B5_Cand2, B6_Cand3, and B7_Cand5 that share a motion vector (step S11).

  Next, the merge selection unit 12 performs the processes of steps S12 to S13 for each merge candidate vector. Here, the merge selection means 12 performs processing in the order of merge candidate vectors B3_Cand4, B5_Cand2, B6_Cand3, and B7_Cand5.

  First, the merge selection unit 12 calculates an evaluation cost for the merge candidate vector B3_Cand4 (step S12).

  Then, the merge selection unit 12 stores the evaluation cost as the best merge index of the block B3 in the merge candidate buffer in association with the merge candidate vector B3_Cand4 (step S13).

  Next, the merge selection unit 12 calculates an evaluation cost for the merge candidate vector B5_Cand2 (step S12).

  As for the block B5, the merge candidate vector B5_Cand3 and the evaluation cost are associated with each other and stored in the merge candidate buffer 208 in the merge selection operation of the block B1. Therefore, the merge selection unit 12 compares the evaluation cost calculated for the merge candidate vector B5_Cand2 with the best merge index of the block B5 stored in the merge candidate buffer 124.

  The merge selection means 12 updates the best merge index stored in the merge candidate buffer 124 if the calculated evaluation cost is smaller than the evaluation cost stored in the merge candidate buffer 124. On the other hand, when the calculated evaluation cost is larger than the evaluation cost stored in the merge candidate buffer 124, the merge selection unit 12 does not update the best merge index stored in the merge candidate buffer 124.

  Similarly, the merge selection unit 12 performs the processes of steps S12 to S13 for the merge candidate vectors B6_Cand3 and B7_Cand5. The merge selecting unit 12 processes the merge candidate vector B6_Cand3 in the same manner as the merge candidate vector B5_Cand2. On the other hand, the merge candidate vector B7_Cand5 is processed in the same manner as the merge candidate vector B3_Cand4.

  The processes in steps S11 to S13 are repeated until all the MVs in the blocks B1 to B12 are set as the copy source MV. When the process of setting the MV of the block B12 as the copy source MV is completed, the process according to the flowchart of FIG. 10 is completed.

  As described above, according to the present embodiment, since the predicted images of a plurality of blocks that are copied from the same motion vector as merge candidate vectors are collectively generated, intermediate data generation or memory access in the decimal pixel interpolation processing is performed. Can be reduced. That is, according to the present embodiment, it is possible to reduce the amount of calculation for creating a predicted image and memory access in the cost evaluation of merge candidate vectors. As a result, according to the present embodiment, the processing time of the motion vector determination process can be reduced.

  According to the present embodiment, by continuously generating predicted images across a plurality of encoding target blocks, it is possible to efficiently determine a motion vector by reducing overlapping processes and memory accesses between blocks.

(Related technology)
Here, related technologies will be described with reference to the drawings. FIG. 11 is a block diagram showing a configuration of a motion vector determination unit 1000 which is a general HEVC reference encoder HM (HEVC Test Model).

  The motion vector determination unit 1000 is different from the motion vector determination unit 10 of the first embodiment in that the merge selection unit 120 does not include the merge candidate buffer 124. The configuration of the motion vector determination unit 1000 is the same as the configuration of the motion vector determination unit 10 of the first embodiment except for the merge candidate buffer 124.

  FIG. 12 is a flowchart for explaining the operation in the merge selection means 120 of the reference encoder HM of the related art.

  The merge selection process by the merge selection unit 120 is sequentially performed on each encoding target block. For example, when a frame divided into blocks B1 to B12 as shown in FIG. 7 is encoded, the upper left block B1 to the lower right block B12 are sequentially encoded as the encoding target blocks. In each encoding target block, the processing of steps S101 to S103 in FIG. 12 is performed on the four merge candidate vectors shown in FIG. 4 in the order of Cand2, Cand3, Cand4, and Cand5, for example.

  In FIG. 12, first, the merge selection means 120 performs the process of block B1. The block B1 is a block located at the upper left corner, and there is no adjacent block on the left side or the upper side, so that the merge selection process is not performed.

  The next block B2 has an adjacent block (block B1) on the left side. Therefore, the merge selection unit 120 performs the processes of steps S12 to S13 for the merge candidate vector Cand4 for copying the motion vector (denoted as B1_MV) of the block B1. The merge candidate vector Cand4 (hereinafter, B2_Cand4) of the block B2 is B1_MV.

  Next, the merge selection unit 120 generates a predicted image of the merge candidate vector B2_Cand4 (step S101). When the merge candidate vector B2_Cand4 (B1_MV) is a decimal precision vector, the merge selection unit 120 generates a prediction image by interpolating decimal pixels.

  Next, the merge selection unit 120 calculates an evaluation cost from the predicted image of the merge candidate vector B2_Cand4 and the input image of the block B2 (step S102).

  Then, the merge selection unit 120 updates the best merge index if the calculated evaluation cost is the smallest among the merge candidate vectors of the block (step S103). Note that the merge candidate vector B2_Cand4 is the merge candidate vector first evaluated for the block B2, so that the evaluation cost of B2_Cand4 is the smallest at that time.

  The merge selecting unit 120 repeats the same processing as that for the block B2 for the blocks B3 and B4.

  In block B5, since there are adjacent blocks on the upper right side and the upper side, merge selection means 120 repeats steps S101 to S103 for Cand2 and Cand3.

  Similarly, the merge selecting unit 120 performs processing for all the encoding target blocks (blocks B1 to B12).

  As described above, in the related art method, the encoding target block loop is the outermost, and the predicted images of B7_Cand4, B9_Cand2, B10_Cand3, and B11_Cand5 are continuous in the decimal interpolation image. Nevertheless, in the related art, when each predicted image is generated, memory access is performed independently, so that efficient memory access is not achieved.

  On the other hand, in the method of the present embodiment, the loop targeted for the MV copied as the merge candidate vector is the outermost, and memory access is collectively performed in the generation of predicted images continuous in the decimal interpolation image. Therefore, according to the present embodiment, efficient memory access is realized.

(Second Embodiment)
Next, a moving picture encoding method according to the second embodiment of the present invention will be described with reference to the drawings.

  In the present embodiment, the technique of the first embodiment is applied to parallel processing using a parallel computer such as a GPU (Graphic Processing Unit). In a GPU program using CUDA (Compute Unified Device Architecture), a large number of threads operate in parallel, but a specific number of threads are grouped into a processing unit called warp. There is a restriction that a plurality of threads included in the same warp always perform the same operation.

  FIG. 13 is a conceptual diagram for explaining a general moving image encoding method. In FIG. 13, the warp is composed of eight threads s1 to s8.

  In step S21 and step S22 in FIG. 13, the same operation is simultaneously processed in all threads in the same warp.

  In the conditional branch in step S23, the threads s1, s2, s4 and s6 and the threads s3, s5, s7 and s8 branch so as to perform different operations. In this case, the operations of threads s1, s2, s4, and s6 are performed in step S24, and the operations of threads s3, s5, s7, and s8 are performed in step S25. In other words, when threads in the same warp perform different calculations, the calculation in step S24 and the calculation in step S25 are sequentially executed. That is, in a general moving image encoding method, sufficient parallel processing performance cannot be obtained unless all threads in the warp perform the same operation.

  When merge index selection is performed in parallel using a GPU or the like, it is common to use a block to be encoded, which is a loop of FIG. 12, and a merge candidate vector as a unit of parallel processing. That is, it is general that each warp processes a different encoding target block, and each thread in the warp processes a separate merge candidate vector.

  For example, in the example of FIG. 7, it is assumed that a certain warp is in charge of the block B6, and a thread in the warp processes merge candidate vectors B6_Cand2, B6_Cand3, B6_Cand4, and B6_Cand5 in parallel. At this time, the copy source vectors of the merge candidate vectors are B3_MV, B2_MV, B5_MV, and B1_MV. As described above, in the case where the predicted image generation process of the merge candidate vector is different, the efficiency of the parallel processing is lowered.

  For example, when motion vectors B3_MV and B2_MV are unidirectional prediction and B5_MV and B1_MV are bidirectional prediction, the process of generating a predicted image is different between unidirectional prediction and bidirectional prediction. Therefore, when unidirectional prediction and bidirectional prediction are processed in parallel with the same warp, both the unidirectional prediction process and the bidirectional prediction process are executed as in the branch process in step S23 of FIG. The amount of computation increases.

  FIG. 14 is a flowchart showing the operation of the moving picture coding method of the present embodiment. The flowchart of FIG. 14 is the same as the flowchart of FIG. 10 except that the copy source MV and the merge candidate vector loop are processed in parallel, and thus detailed description thereof is omitted.

  In the present embodiment, the operation of the first embodiment is applied to parallel processing. That is, in this embodiment, as shown in FIG. 14, a loop of the copy source MV and the merge candidate vector is set as a parallel processing unit. That is, in this embodiment, each warp processes a different copy source MV, and each thread in the warp processes a separate merge candidate vector.

  For example, a certain warp takes charge of the motion vector B6_MV of the block B6 as the copy source MV, and a thread in the warp processes the merge candidate vectors B9_Cand2, B10_Cand3, B11_Cand5, and B7_Cand4 in parallel. Since these merge candidate vectors are all copies of B6_MV, the predicted image creation processes of B9_Cand2, B10_Cand3, B11_Cand5, and B7_Cand4 are all the same process.

  That is, the dynamic encoding device of this embodiment includes a plurality of merge selection means 12 (also referred to as selection means) (FIG. 5). The plurality of merge selection means 12 are assigned to a plurality of arithmetic devices operating in cooperation with each other. The plurality of merge selection units 12 are assigned calculation of evaluation costs associated with any of the plurality of encoding target blocks, and a plurality of merge vectors are duplicated with a motion vector associated with any of the encoding target blocks. A prediction image of the encoding target block is generated in parallel.

  According to the method of the present embodiment, since all the threads in the warp perform the same operation, it is possible to extract sufficient parallel processing performance.

  Furthermore, according to the present embodiment, it is possible to simultaneously reduce the memory access between blocks and the duplication of intermediate data generation, which are the effects of the sequential processing in the first embodiment.

  In this embodiment, one candidate vector is processed by one thread and one copy source MV is processed by one warp. However, the correspondence between the thread and the process is not limited to this. The method of the present embodiment is sufficient if there is only one copy source MV of merge candidate vectors that are simultaneously processed by a plurality of threads in one warp, and may have a variation such that one candidate vector is processed by a plurality of threads.

(hardware)
Here, the hardware 90 for realizing the moving picture coding apparatus according to the embodiment of the present invention will be described with reference to FIG. Note that the hardware 90 is an example for realizing the moving image encoding apparatus of the present embodiment, and does not limit the scope of the present invention.

  As shown in FIG. 15, the hardware 90 includes a processor 91, a main storage device 92, an auxiliary storage device 93, an input / output interface 95, and a network adapter 96. The processor 91, main storage device 92, auxiliary storage device 93, input / output interface 95, and network adapter 96 are connected to each other via a bus 99. The processor 91, the main storage device 92, the auxiliary storage device 93, and the input / output interface 95 are connected to a network such as an intranet or the Internet via a network adapter 96. Note that each of the components of the hardware 90 may be single or plural.

  The processor 91 is an arithmetic device that expands a program stored in the auxiliary storage device 93 or the like in the main storage device 92 and executes the expanded program. In the present embodiment, a configuration using a software program installed in the hardware 90 may be used. The processor 91 executes various arithmetic processes and control processes. For example, the processor 91 is realized by a CPU (Central Processing Unit). Further, a part of the function of the processor 91 may be imposed on a GPU (Graphics Processing Unit). For example, a configuration including a CPU 911 and a GPU 912 can be adopted as in the hardware 90-2 in FIG.

  The main storage device 92 has an area where the program is expanded. The main storage device 92 may be a volatile memory such as a DRAM (Dynamic Random Access Memory). Further, a nonvolatile memory such as an MRAM (Magnetoresistive Random Access Memory) may be configured and added as the main storage device 92.

  The auxiliary storage device 93 is a storage device for storing various data. The auxiliary storage device 93 is configured as a local disk such as a hard disk or a flash memory.

  The input / output interface 95 is an interface (I / F) that connects the hardware 90 and peripheral devices based on a connection standard.

  If necessary, the hardware 90 may be connected to input devices such as a keyboard, mouse, and touch panel, and output devices such as a display and a printing device. These input devices and output devices can be used for inputting information and settings, outputting moving image information, and the like. Data exchange between the processor 91 and the input device may be mediated by the input / output interface 95.

  The network adapter 96 is an interface for connecting to a network based on standards and specifications. The input / output interface 95 and the network adapter 96 may be shared as an interface for connecting to an external device.

  The hardware 90 may be provided with a reader / writer as necessary. The reader / writer is connected to the bus 99. The reader / writer mediates reading of data programs from the recording medium, writing of processing results of the hardware 90 to the recording medium, and the like between the processor 91 and a recording medium (program recording medium) (not shown). The recording medium can be realized by a semiconductor recording medium such as an SD (Secure Digital) card or a USB (Universal Serial Bus) memory. The recording medium may be realized by a magnetic recording medium such as a flexible disk, an optical recording medium such as a CD (Compact Disc) or a DVD (Digital Versatile Disc), and other recording media.

  It is preferable that the video encoding apparatus according to the embodiment of the present invention can execute parallel processing. FIG. 17 is a configuration example of the GPU 912 that executes parallel processing by the video encoding device according to the present embodiment.

  17 includes an interface 921, a cache 922, and a plurality of streaming multiprocessors 923 (hereinafter, SM: Streaming Multiprocessor). The configuration in FIG. 17 is a conceptual configuration example, and the actual GPU 912 includes components other than the interface 921, the cache 922, and the SM 923.

  The interface 921 is a connection unit for exchanging data with the main storage device 92, the CPU 911, and the like realized by a DRAM or the like.

  The cache 922 is a high-speed memory shared between SMs.

  SM923 is a calculation unit for processing the moving image encoding process. For example, the SM 923 includes a plurality of cores (arithmetic units) called streaming processors, a memory such as a shared memory and a register.

  The above is an example of hardware for realizing the moving picture encoding apparatus according to the embodiment of the present invention. The component of each embodiment of the present invention can be realized as a circuit including at least one of the components included in the hardware of FIGS. Moreover, the component of each embodiment of this invention is realizable as software which operate | moves on the computer which has a hardware structure of FIGS. 15-17.

  Note that the hardware configurations in FIGS. 15 to 17 are examples of a hardware configuration for enabling the moving picture encoding apparatus according to the present embodiment, and do not limit the scope of the present invention. A program that causes a computer to execute processing by the moving image encoding apparatus of the present embodiment is also included in the scope of the present invention. Furthermore, a program recording medium that records the program according to the embodiment of the present invention is also included in the scope of the present invention.

  Although the present invention has been described with reference to the embodiments, the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  According to the present invention, it is possible to reduce the amount of calculation and memory access in moving picture encoding, so that high-speed processing of high-resolution video can be realized. For example, the present invention can be used for a moving image encoding system, an imaging system, a transcoding system, and the like that require high resolution processing.

DESCRIPTION OF SYMBOLS 1 Moving image encoding apparatus 10 Motion vector determination means 11 Motion search means 12 Merge selection means 13 Motion vector comparison means 14 Motion vector buffer 21 Inter prediction means 22 Intra prediction means 23 Adder 24 Conversion / quantization means 25 Encoding means 26 Inverse transformation / inverse quantization means 27 Synthesizer 28 Loop filter 29 Frame buffer 120 Merge selection means 121 Evaluation cost calculation means 122 Merge index selection means 123 Merge candidate predicted image generation means 124 Merge candidate buffer

Claims (10)

The reference from the first buffer that receives the encoding target block included in the encoding target frame and the reference image corresponding to the encoding target block and stores the motion information associated with the encoded block A predicted image generating unit that acquires the motion information candidate associated with an image and generates a predicted image using the motion information candidate and the reference image;
An evaluation cost calculating unit that receives the encoding target block and the prediction image, and calculates an evaluation cost of the encoding target block using the encoding target block and the prediction image;
A second buffer that stores the motion information and the evaluation cost in association with each other using the motion information as a candidate to be associated with the encoding target block;
The index stored in the second buffer is updated based on the evaluation cost calculated with respect to the motion information, and the motion information that is a candidate to be associated with the encoding target block is updated based on the evaluation cost. An encoding device comprising selection means having index selection means for selecting from two buffers. The first buffer;
The selection means;
Search means for receiving the encoding target block and the reference image as input and determining the motion information to be associated with the encoding target block in a mode for directly encoding the motion information;
Comparing means for comparing the motion information determined by the search means with the motion information selected by the selection means based on the evaluation cost with respect to the encoding target block, and selecting any of the motion information The encoding apparatus according to claim 1, further comprising motion information determination means having The index selection means includes
If the evaluation cost calculated for the motion information candidate associated with the block to be encoded is smaller than the evaluation cost of the motion information stored in the second buffer, the evaluation cost is stored in the second buffer. The encoding apparatus according to claim 2, wherein the index is updated. The predicted image generation means includes
The encoding apparatus according to claim 2 or 3, wherein the prediction images corresponding to the plurality of encoding target blocks having the common motion information as candidates are collectively generated. A frame buffer for storing the reference image corresponding to the plurality of encoding target blocks constituting the encoding target frame;
The motion information determined by the motion information determination means is input, the reference image corresponding to the motion information is read from the frame buffer, and the prediction image of the encoding target block corresponding to the motion information is generated. Inter prediction means,
An adder that calculates a residual signal that is a difference between the encoding target block and the predicted image;
Transform / quantization means for inputting the residual signal from the adder and generating a transform coefficient quantized by integer transforming the residual signal;
An inverse transform / inverse quantization unit that inputs the quantized transform coefficient from the transform / quantization unit, and inversely transforms the quantized transform coefficient to generate a prediction residual image;
The prediction residual image is input from the transform / inverse quantization unit, the prediction image is input from the inter prediction unit, and the reconstructed image is generated by synthesizing the prediction residual image and the prediction image. A synthesizer;
Intra prediction means for inputting the reconstructed image from the synthesizer and generating the predicted image of the coding target block from the reconstructed image based on intra prediction;
A loop filter that inputs the reconstructed image from the combiner, removes block distortion of the reconstructed image, and stores the reconstructed image from which block distortion has been removed as the reference image in the frame buffer;
5. The encoding unit according to claim 2, further comprising: an encoding unit that inputs the quantized transform coefficient from the transform / quantization unit and outputs a bitstream obtained by encoding the quantized transform coefficient. The encoding device described in 1. A plurality of the selection means;
The plurality of selection means are assigned to a plurality of arithmetic devices that operate in cooperation with each other, assigned to calculate the evaluation cost associated with any one of the plurality of encoding target blocks, and any one of the codes The encoding apparatus according to any one of claims 1 to 5, wherein the prediction images of the plurality of encoding target blocks that duplicate the motion information associated with the encoding target block are generated in parallel. An encoding target block included in the encoding target frame and a reference image corresponding to the encoding target block are input,
Obtaining the motion information associated with the reference image from a first buffer in which motion information associated with the encoded block is stored;
Generating a predicted image using the reference image and the motion information;
An evaluation cost is calculated for the encoding target block,
Using the motion information that is a candidate to be associated with the encoding target block as an index, the motion information and the evaluation cost are associated with each other and stored in a second buffer,
Updating the index stored in the second buffer based on the evaluation cost calculated for the motion information;
The encoding method which selects the said motion information used as the candidate matched with the said encoding object block from a said 2nd buffer based on the said evaluation cost. Selecting the motion information of the encoding target block as a replication source;
Generating the predicted images of a plurality of the encoding target blocks with the motion information selected as a copy source as candidates,
Calculating the evaluation cost for the motion information that is a candidate to be associated with the encoding target block;
When the evaluation cost calculated for the motion information that is a candidate to be associated with the encoding target block is smaller than the evaluation cost stored in the second buffer, the index stored in the second buffer The encoding method according to claim 7, wherein the encoding is updated. A process of inputting an encoding target block included in the encoding target frame and a reference image corresponding to the encoding target block;
Processing for obtaining the motion information associated with the reference image from a first buffer in which motion information associated with the encoded block is stored;
A process of generating a predicted image using the reference image and the motion information;
Processing for calculating an evaluation cost for the encoding target block;
A process of storing the motion information and the evaluation cost in association with each other using the motion information that is a candidate to be associated with the encoding target block as an index;
Updating the index stored in the second buffer based on the evaluation cost calculated for the motion information;
An encoding program that causes a computer to execute processing for selecting the motion information that is a candidate to be associated with the encoding target block from the second buffer based on the evaluation cost. A process of selecting the motion information of the encoding target block as a replication source;
A process of collectively generating the predicted images of a plurality of the encoding target blocks having the motion information selected as a replication source as candidates;
A process of calculating the evaluation cost for the motion information that is a candidate to be associated with the encoding target block;
When the evaluation cost calculated for the motion information that is a candidate to be associated with the encoding target block is smaller than the evaluation cost stored in the second buffer, the index stored in the second buffer The encoding program according to claim 9, which causes a computer to execute a process for updating the data.

Images