JP4395174B2 - Moving picture coding method, moving picture parallel coding encoder, moving picture parallel coding method, moving picture parallel coding apparatus, program thereof, and computer-readable recording medium recording the program - Google Patents

Description

  The present invention relates to a moving image parallel coding method for performing moving image coding in parallel using a plurality of encoders.

  When parallel encoding processing is performed in accordance with the MPEG-2 standard (ISO / IEC 13818-2) [see non-patent document 1], each frame of the input moving image is divided into N regions by horizontally dividing the frame. A method has been proposed in which a strip-shaped divided region is generated, and N encoders are in charge of encoding each strip-shaped divided region [see Non-Patent Document 2].

  FIG. 9 is a block diagram showing an example of parallel encoding processing conforming to the conventional MPEG-2 standard. FIG. 10 shows an example in which the input frame is divided into strip-shaped regions. FIG. 9 shows an example in which an input frame is divided into four strip-shaped areas as shown in FIG. 10 and parallel encoding operations are performed by four encoders 102 to 105.

  The input frame is input to the screen division processing unit 101, and is divided into N (N = 4 in this example) strip-like regions (1 to 4). Each of the generated strip-shaped regions is input to the encoders 102 to 105 and is encoded simultaneously in parallel, and the divided encoded data is output from each of the encoders 102 to 105, respectively. These divided encoded data are input to the stream concatenation unit 106. Since each piece of divided encoded data is in a state in which byte alignment is achieved, encoded data that conforms to the MPEG-2 standard is output by simply connecting them at the stream connecting unit 106.

  The encoders 102 to 105 compliant with the MPEG-2 standard have the same configuration. A configuration example of the encoders 102 to 105 is shown in FIG.

  Each of the encoders 102 to 105 uses a strip-shaped area as an input image to be processed. The input strip-shaped area is input to the macroblock dividing unit 110. The macroblock dividing unit 110 divides the strip-like area into 16 × 16 pixel (in the case of luminance signals) encoding processing units called macroblocks, and outputs them to the subtractor 111 as processing target macroblocks.

  At the same time as the input of the processing target macroblock, an inter-predicted image macroblock is input to the subtractor 111 when the intra / inter selection unit 112 selects inter. When an intra is selected, an invalid image (pixel) is selected. Macroblocks whose values are all 0) are input as predicted macroblocks. The subtractor 111 outputs a result obtained by subtracting them in units of pixels in the macroblock to the DCT unit 113.

  The DCT unit 113 performs DCT conversion on the output of the subtractor 111 and outputs a DCT conversion coefficient to the quantization unit 114. The quantization unit 114 performs a quantization process at the designated quantization step, and outputs a quantization coefficient to the information source encoding unit 115. The information source encoding unit 115 performs two-dimensional variable length encoding using a Huffman code, and outputs divided encoded data.

  The quantization coefficient output from the quantization unit 114 is also input to the inverse quantization unit 116, where inverse quantization processing is performed, and the inverse quantized DCT transform coefficient is output from the inverse quantization unit 116. The inverse quantized DCT transform coefficient is input to the inverse DCT unit 117, and the transformed macroblock subjected to the inverse DCT transform by the inverse DCT unit 117 is output to the adder 118. The adder 118 is also input with the prediction macroblock, the addition process is performed for each pixel, and the local decoding macroblock as the addition result is output to the frame memory 119.

  In the frame memory 119, macroblocks are reconstructed in the order of horizontal raster scan, and are recorded as strip-shaped area local decoded images. It should be noted that the strip-shaped area local decoded images are respectively transferred between the adjacent encoders 102 to 105, thereby acquiring the strip-shaped area local decoded images owned by the adjacent encoders. The local decode image of the entire input frame may be generated by combining the strip-shaped area local decode images and recorded in the frame memory 119.

  The motion prediction unit 120 receives the processing target macroblock and the local decoded image, performs a motion estimation process, and outputs the estimated motion vector to the motion compensation unit 121. The motion compensation unit 121 extracts a macroblock at a position indicated by the motion vector in the local decoded image, and outputs the macroblock to the intra / inter selection unit 112 as an inter predicted image macroblock.

  Similarly, H.M. The H.264 standard (ISO / IEC 14496-10) [see Non-Patent Document 3] also performs parallel encoding processing in accordance with H.264 standard. When the function of the loop filter process (deblocking filter process), which is a feature of the H.264 standard, is turned off, it is possible to realize the moving image parallel encoding process with the same configuration as in FIG.

  That is, in the encoder in FIG. 9, the input frame is input to the screen division processing unit 101 as in the case of performing the encoding process in conformity with the MPEG-2 standard, and N strips as shown in FIG. It is divided into like areas. Each of the generated strip-shaped regions is input to the encoders 102 to 105, respectively, and at the same time in parallel with H.264. The encoded data is encoded in accordance with the H.264 standard, and divided encoded data is output from each of the encoders 102 to 105. The divided encoded data output from the encoders 102 to 105 is input to the stream concatenation unit 106. Since each piece of divided encoded data is in a state in which the byte alignment is taken, the stream concatenation unit 106 simply concatenates each of the encoded data. The encoded data conforming to the H.264 standard is output. H. FIG. 12 shows a configuration example of the encoders 102 to 105 in the case of performing parallel encoding in accordance with the H.264 standard (loop filter setting is OFF).

  Each H. The encoders 102 to 105 that encode the strip-shaped area in accordance with the H.264 standard use the strip-shaped area as an input image to be processed. This strip area is input to the macroblock dividing unit 210. The macroblock dividing unit 210 divides the strip-like area into coding processing units called macroblocks, and outputs them to the subtractor 211 as processing target macroblocks.

  When the inter is selected by the intra / inter selection unit 212, the inter prediction image macro block is input to the subtractor 211 at the same time as the processing target macro block. When the intra is selected, the intra prediction image macro block is input. Input as a prediction macroblock. The subtractor 211 outputs a result obtained by subtracting them in units of pixels in the macroblock to the integer DCT unit 213.

  The integer DCT unit 213 performs DCT conversion on the output of the subtractor 211 and outputs a DCT conversion coefficient to the quantization unit 214. The quantization unit 214 performs quantization processing at the designated quantization step, and outputs the quantization coefficient to the information source encoding unit 215. The information source coding unit 215 performs information source coding processing using a context adaptive variable length code called CAVLC or a context adaptive arithmetic code called CABAC, and outputs divided coded data.

  The quantization coefficient output from the quantization unit 214 is also input to the inverse quantization unit 216, where the inverse quantization unit 216 performs an inverse quantization process, and outputs an inverse-quantized DCT transform coefficient. The inverse-quantized DCT transform coefficient is input to the inverse integer DCT unit 217, subjected to inverse integer DCT processing, and the converted macroblock is output to the adder 218. The adder 218 also receives the prediction macroblock, performs addition processing for each pixel, and outputs the local decode macroblock to the pre-filter frame memory 219. In the pre-filter frame memory 219, the macroblocks are reconstructed in the raster scan order and recorded as a pre-filter strip area local decoded image.

  The pre-filter strip area local decoded image is input to the loop filter unit 220 when the loop filter is ON. The loop filter unit 220 removes block noise at the macroblock boundary from the input image, outputs it to the post-filter frame memory 221 as a post-filter strip region local decoded image, and accumulates it in the post-filter frame memory 221. As an operation of turning off the loop filter, the pre-filter strip region local decode image and the post-filter strip region local decode image are the same, and the image stored in the pre-filter frame memory 219 is used as it is. It is transferred to the memory 221 and stored.

  The parallel coding process can be performed by dividing the screen into horizontal strips with the configuration shown in FIGS. 9 and 12 only when the loop filter is selected OFF. When the loop filter is ON, there is a problem that will be described later. Therefore, in the configuration shown in FIGS. It is impossible to realize encoding conforming to the H.264 standard.

  Similarly to the configuration shown in FIG. 9, the filtered strip-shaped area local decoded images are transferred between the adjacent encoders 102 to 105, respectively, so that the filtered strip-shaped area locals owned by the adjacent encoders are transferred. The decoded image may be acquired, and the filtered local decoded image of the entire input frame may be generated by combining the upper and lower adjacent strip-shaped area local decoded images and recorded in the filtered frame memory 221. By recording together the upper and lower adjacent strip-shaped area local decoded images in the post-filter frame memory 221, it becomes possible to extend the search range of motion prediction / motion compensation processing described later to outside the strip-shaped area. Coding efficiency is improved.

  The motion prediction unit 222 receives the processing target macroblock and the filtered local decoded image, performs a motion estimation process, and outputs the estimated motion vector to the motion compensation unit 223. The motion compensation unit 223 extracts a macroblock whose position is indicated by the motion vector in the filtered local decoded image, and outputs the macroblock to the intra / inter selection unit 212 as an inter predicted image macroblock.

  On the other hand, the pre-filter strip area local decoded image is input from the pre-filter frame memory 219 to the intra prediction mode evaluation unit 224 and the intra prediction value generation unit 225. H. In the H.264 standard, intra prediction is performed on a local decoded image before the processing in the loop filter unit 220 is performed, and thus the configuration is as described above. The intra prediction mode evaluation unit 224 selects an appropriate prediction mode from various selectable intra prediction modes, and outputs an intra prediction mode code to the intra prediction value generation unit 225. The intra-prediction value generation unit 225 generates an intra-prediction image macroblock from the pre-filter strip region local decoded image and the intra-prediction mode code, and outputs the intra-prediction image macroblock to the intra-inter selection unit 212.

H. In the H.264 standard, it is possible to enable / disable the loop filter by using a flag called disable_deblocking_filter_idc existing in the slice header.
ITU-T Recommendation H.262 (2000) | ISO / IEC 13818-2: 2000, “Information technology-Generic coding of moving pictures and associated audio information: Video”, International Standard, second edition, December 2000. Nakamura et al., “MPEG-2 HDTV Encoder System Using Multiple SDTV Encoder LSIs”, IEICE, IEICE Technical Report, ICD2001-78 (2001-08), pp.91-96. ITU-T Recommendation H.264: “Advanced Video Coding for generic audiovisual services” (2003) | ISO / IEC 14496-10: “Coding of audiovisual objects-Part 10: Advanced Video Coding” (2003).

  H. The H.264 standard stipulates that an image obtained by performing filter processing on a local decoded image is used for subsequent inter prediction image generation. In the loop filter processing, as shown in FIG. 13, at least the decoder performs filter processing for each macroblock in the horizontal raster scan order from the upper left macroblock to the lower right macroblock in the decoder. Is stipulated. Therefore, both the left and upper adjacent macroblocks of the filtering target macroblock need to be filtered, and the filtering is performed between both of the macroblocks that have already been filtered and the encoding target macroblock. Processing needs to be done.

  A configuration example of the loop filter unit 220 is shown in FIG. The pre-filter strip region local decoded image input to the loop filter unit 220 is input to the macroblock dividing unit 301. The macroblock dividing unit 301 divides the strip-like area into encoding processing units called macroblocks, and outputs each macroblock as a processing target macroblock to the vertical edge horizontal filter unit 302 in the order of horizontal raster scanning. The vertical edge horizontal filter unit 302 sequentially performs a filtering process in the horizontal direction on every vertical edge in the macroblock.

  The processing outline of the vertical edge horizontal filter unit 302 is shown on the left side of FIG. The vertical edge horizontal filter unit a303 performs a filter process of 8 horizontal pixels × 1 vertical pixel centering on each pixel constituting the vertical edge a. Here, filter processing is performed using the pixel value of the macroblock and the pixel value of the macroblock that has been subjected to the filter processing on the left as input pixel values. The pixel value resulting from the filtering process on the vertical edge a is subjected to the filtering process on the vertical edge b in the vertical edge horizontal filter unit b304. Similarly, as shown in FIG. 16, the horizontal filter processing for the vertical edge c is performed on the horizontal filter processing result for the vertical edge b, and the horizontal filter processing for the vertical edge c is performed in the vertical edge horizontal filter section c305. The horizontal filter processing for the vertical edge d is performed in the vertical edge horizontal filter unit d306.

  The halfway filter macroblock that has been subjected to the horizontal filter processing as described above is input to the horizontal edge vertical filter unit 307. The horizontal edge vertical filter unit 307 sequentially performs filtering in the vertical direction on every horizontal edge in every four pixels in the macroblock. The processing outline of the horizontal edge vertical filter unit 307 is shown on the right side of FIG.

  The horizontal edge vertical filter unit e308 performs a filter process of 1 horizontal pixel × 8 vertical pixels centering on each pixel constituting the horizontal edge e. Here, processing is performed using both the pixel value of the macroblock and the pixel value of the macroblock that has been subjected to the upper adjacent filter processing as input signal values. The pixel value resulting from the filtering process on the horizontal edge e is subjected to the filtering process on the horizontal edge f in the horizontal edge vertical filter unit f309. Similarly, as shown in FIG. 16, the vertical filter processing for the horizontal edge g is performed on the vertical filter processing result for the horizontal edge f in the vertical filter unit g310 for the vertical filter processing result for the horizontal edge f. The horizontal edge vertical filter unit h311 performs vertical filter processing on the horizontal edge h.

  As indicated above, H.C. Since the loop filter processing in the H.264 standard is configured to sequentially perform filter processing while reusing filtered pixels, it is necessary to appropriately perform the order of filter processing. In addition, in order to perform filter processing of a certain macroblock, it is necessary that filter processing has already been completed for the macroblocks on the upper and left sides.

  The post-filter macroblock obtained by performing the filtering process with respect to the horizontal edge / vertical filter unit h311 is input to the post-filter macroblock memory 312 and reconstructed in the order of horizontal raster scanning, and the post-filter strip-like region local decoded image As recorded.

  FIG. It is a figure explaining the problem in a H.264 parallel encoding process. Consider a case where the moving image parallel encoding processing configuration shown in FIGS. 9 and 12 is used and the loop filter processing is enabled. Each loop filter unit 220 in the encoders 102 to 105 performs filter processing in the order of horizontal raster scanning from the upper left macroblock to the lower right macroblock in each strip-shaped region as shown in FIG. It will be.

  When considering the parallel processing macroblocks in the strip-like areas 2, 3 and 4 in FIG. 17, the upper adjacent macroblock is in an unfiltered state when performing the processing of the horizontal edge vertical filter unit e308. I understand. For this reason, the vertical filter processing performed by the horizontal edge vertical filter unit e308 is H.264. H.264 standard, resulting in H.264 filter processing. A filtered strip region local decoded image different from the decoded image obtained by the decoder compliant with H.264 is generated. This causes a phenomenon of tailing in the video due to the mismatch between the decoded video of the encoder and the decoder, which is called drift, and thus causes unacceptable distortion.

  As described above, the loop filter can be set to OFF (disable_deblocking_filter_idc = 1). However, in this case, the encoding efficiency is remarkably deteriorated, and the image quality such as block distortion is deteriorated particularly at the low rate encoding. Further, in order to be able to use the function of the loop filter processing even in the moving image parallel coding configuration using the configurations as shown in FIGS. 9 and 12, the loop filter is turned off only at the border of the rectangular rectangular region, and other adjacent regions are also used. A mode setting (disable_deblocking_filter_idc = 2) that allows the function of the loop filter processing to be used only at the boundary of the macroblock to be performed is also described in H.264. Although it is prepared in the H.264 standard, a decoded image in which a joint is conspicuous only at the upper and lower adjacent borders of the strip-shaped region is generated, which causes visual disturbance.

  As mentioned above, H.C. When the parallel coding is performed in accordance with the H.264 standard, the loop filter function is enabled, and the screen is divided into horizontal strip-like areas, there is a mismatch between the decoded images between the encoder and the decoder. , Distortion such as drift occurs. In order to avoid these problems, it is possible to disable the loop filter function or disable only between the strip-shaped regions. Compared with H.264, encoding efficiency is deteriorated.

  The present invention has been made to solve the above problems. The H.264 standard (including substantially equivalent standards) remains completely effective while the loop filter function defined in the H.264 standard is valid. It is an object to enable generation of encoded data that conforms to the H.264 standard.

  In order to solve the above problem, in the present invention, for example, H.264. H.264 standard is defined, and the loop filter function is enabled, and a method of performing parallel coding by dividing the screen into horizontal strip-like regions is defined.

  The moving image encoding method of the present invention is a moving image encoding method in which each frame of an input moving image is divided into rectangular small regions in a grid pattern, and the rectangular small regions are sequentially encoded in the raster scan order in the horizontal direction. A local decoded image that is the same as the decoded image obtained on the decoder side for the already encoded frame is generated on the encoder side, and an image obtained by filtering the local decoded image is used as a reference image. In the moving picture coding method that performs interframe predictive coding of subsequent frames using as a rectangular small area obtained by dividing an input moving picture frame in a grid pattern, Encoding for each rectangular small area toward the area in the order of horizontal raster scan, and for the rectangular small area obtained by dividing the local decoded image into a grid, Characterized by the small rectangular area in each small rectangular area toward the small rectangular area of the bottom right and a step of performing a filtering process in the vertical raster scan order.

  In the moving picture encoding method, a vertical raster for each rectangular small area from a rectangular small area in the upper left to a rectangular small area in the lower right is obtained for the rectangular small areas obtained by dividing the local decoded image in a grid pattern. When performing the filtering process in the scan order, as the last filtering step of the filtering process corresponding to the rectangular small area, an adjacent boundary pixel between the rectangular small area and the rectangular small area adjacent to the right of the rectangular small area is set. By performing the filter processing, The present invention is characterized in that encoded data compliant with the H.264 standard (ITU-T H.264) is generated.

  Further, in the moving image encoding method, after encoding of all rectangular small areas obtained by dividing the input moving image frame in a grid pattern in the horizontal raster scan order is completed, the local decoded image is divided into a grid pattern. The filtering process is started in the order of the vertical raster scan for all the rectangular small areas obtained in this way.

  The moving image parallel coding encoder of the present invention generates a plurality of strip-shaped divided regions by horizontally dividing each frame of an input moving image, and each of the plurality of encoders has a strip shape. Coding for encoding the strip-shaped divided region used in a moving image parallel coding apparatus that divides the divided region into rectangular small regions in a grid pattern and encodes the rectangular small regions in the strip-shaped divided region in parallel. A rectangular small region obtained by dividing the strip-shaped divided region into a lattice shape, from the upper left rectangular small region in the strip-shaped divided region to the lower right rectangular small region in the strip-shaped divided region Means for encoding each rectangular small area in the order of horizontal raster scan, and dividing the strip-shaped divided areas of the locally decoded image obtained by decoding the encoded data of the encoded strip-shaped divided areas into a grid pattern. A small rectangular area Means for performing a filtering process in the order of vertical raster scan for each rectangular small area from the upper left rectangular small area in the rectangular divided area toward the lower right rectangular small area in the rectangular divided area; If there is an encoder in charge of encoding the strip-shaped divided area adjacent to the upper side of the rectangular divided area, a rectangle adjacent to at least the uppermost rectangular small area of the strip-shaped divided area from the encoder If there is a means for receiving the image after filtering of the small area and recording it in the memory, and an encoder in charge of encoding the adjacent strip-shaped divided area below the strip-shaped divided area, the code Means for transferring a filtered image of at least the rectangular small area at the bottom of the strip-shaped divided area toward the generator, and the means for performing the filtering process has an uppermost portion of the strip-shaped divided area. For a small rectangular area When performing the filtering process, the filtering process is performed using the filtered image recorded in the memory, and the encoding means uses the filtered local decoded image as a reference image to generate a subsequent frame. Inter-frame predictive coding is performed.

  Further, in the moving image parallel encoding encoder, the filter processing means targets pixels in the rectangular small area in the filtering process of one rectangular small area in the strip-shaped divided area. After performing a horizontal filtering process on the vertical edge and a vertical filtering process on the horizontal edge, an adjacent boundary pixel between the rectangular small area and the rectangular small area adjacent to the right of the rectangular small area is determined. A horizontal filtering process is performed on a vertical edge as a target.

  The moving image parallel coding method according to the present invention generates a plurality of strip-shaped divided regions by horizontally dividing each frame of an input moving image into a plurality of strip-shaped divided regions. A moving image parallel encoding method for dividing a rectangular small region in a strip-shaped divided region into a grid shape and sequentially encoding the rectangular small regions in a raster scan order in the horizontal direction, and the already-encoded strip-shaped divided region A local decoded image that is the same image as the decoded image obtained on the decoder side is generated on each encoder side, and an image obtained by filtering the strip-shaped divided region at the same position in the local decoded image is a reference image. In the moving picture parallel coding method for performing interframe predictive coding of subsequent frames by using as a rectangular small area obtained by dividing the strip-like divided area into a grid shape, each encoder has the strip-like shape. Split Encoding in the order of horizontal raster scan for each rectangular small area from the upper left rectangular small area to the lower right rectangular small area in the strip-shaped divided area, and a strip at the same position in the local decoded image For each rectangular small area obtained by dividing the rectangular divided area into a grid pattern, each encoder changes from the rectangular upper left area in the rectangular divided area to the lower right rectangular small area in the rectangular divided area. And performing a filtering process in the order of vertical raster scanning for each rectangular small area.

  Further, in the moving image parallel coding method, for each rectangular small region obtained by dividing the strip-shaped divided region in the local decoded image into a grid shape, each encoder has a rectangular shape at the upper left in the strip-shaped divided region. When performing filter processing in the order of vertical raster scan for each rectangular small area from the small area toward the lower right rectangular small area in the strip-shaped divided area, the last filter of the filter processing corresponding to the rectangular small area As a processing step, a filter process is performed on an adjacent boundary pixel between the rectangular small region and the rectangular small region adjacent to the right of the rectangular small region. The present invention is characterized in that encoded data compliant with the H.264 standard (ITU-T H.264) is generated.

  Further, in the moving image parallel coding method, each encoder performs coding in the horizontal raster scan order of all rectangular small regions obtained by dividing each strip-like divided region of the input moving image frame into a grid pattern. After completion, the filtering process is started in the vertical raster scan order for all rectangular small areas obtained by dividing the strip-like divided areas at the same position in the local decoded image into a grid pattern.

  In the moving image parallel encoding method, a certain encoder performs a filtering process on the C-th rectangular small region from the top of the rectangular small region group in the B-th column from the left of the A-th strip-shaped divided region from the top. In parallel with this, another encoder performs a filtering process on the C-th rectangular small region from the top of the (A + B−W) -th rectangular small region group from the left of the W-th strip-shaped divided region from the top. It is characterized by that.

  In the moving image parallel coding method, the encoder in charge of the Xth strip-shaped divided region from the top is a vertical raster of the rectangular small region group in the Yth column from the left of the Xth strip-shaped divided region. A step of performing a filtering process in the scan order and an Xth strip shape from an encoder in charge of the Xth strip-shaped divided region to an encoder in charge of the (X + 1) th strip-shaped divided region A step of transferring the filtered image of the bottom rectangular small area of the Yth column from the left of the divided area, and an encoder in charge of the (X + 1) th strip-shaped divided area, Receiving the filtered image of the lowermost rectangular small area of the Yth column from the left of the strip-shaped divided area and recording it in the memory, and the encoding responsible for the (X + 1) th strip-shaped divided area The filter of the received rectangular small area Using an image after processing, and executes the steps of performing (X + 1) -th most filtering small rectangular area above the Y-th column from the left of the strip-shaped divided areas.

  The moving image parallel coding apparatus of the present invention uses each of the plurality of moving image parallel coding encoders to form each strip-shaped divided region obtained by dividing each input frame by each moving image parallel coding encoder. The rectangular small areas in the strip-shaped divided areas are encoded in order in a raster scan order in the horizontal direction by dividing into rectangular small areas in a grid pattern.

[Action]
By using the present invention, an input frame is divided into horizontal strip-like divided regions (hereinafter referred to as strip-like regions). By performing H.264 parallel coding, the H.264 decoding is completely performed without any mismatch between decoded images between the encoder and the decoder. It is possible to generate encoded data that conforms to the H.264 standard. Furthermore, the parallelism can be easily increased by increasing the number of horizontal strip divisions, and a large-screen encoding device using a plurality of lower scale LSIs can be developed.

  H. Since the parallel coding process can be performed while the loop filter function defined in the H.264 standard is enabled, there is no deterioration in image quality caused by disabling the loop filter process in order to perform the parallel coding process. .

  As described above, the present invention is used to divide the input frame into horizontal strip-like areas and to generate H.264. By performing H.264 parallel coding (including a standard substantially equivalent to the H.264 standard), there is no discrepancy between decoded images between the encoder and the decoder. It is possible to generate encoded data that conforms to the H.264 standard.

  H. Since the parallel coding process can be performed while the loop filter function defined in the H.264 standard is enabled, there is no deterioration in image quality caused by disabling the loop filter process in order to perform the parallel coding process. .

  Examples of the present invention will be described below. FIG. 1 shows the overall configuration of a moving image parallel encoding apparatus including the present invention. In the overall configuration diagram shown in FIG. 1, the input frame is divided into four strip-shaped areas, and four H.264 frames are divided. An example in which a parallel encoding operation is performed by an H.264 encoder is described.

  In FIG. 1, an input frame is input to a screen division processing unit 21 and divided into N (N = 4 in this example) strip-like regions. An example in which the input frame is divided into strip-like regions is the same as in FIG. Each of the generated strip-shaped regions is H.264. H.264 encoders 22 to 25, which are simultaneously encoded in parallel. H.264 encoders 22 to 25 respectively output divided encoded data. These divided encoded data are input to the stream concatenation unit 26. Since each piece of divided encoded data is in a state in which the byte alignment is taken, the stream concatenation unit 26 simply concatenates each of the H.264 data. The encoded data conforming to the H.264 standard is output. Hereinafter, the macroblock may be abbreviated as “MB”.

  This H. In the H.264 encoders 22 to 25, the H.264 encoder In performing loop filter processing in H.264 encoding, H.264 The H.264 encoder 22 is a H.264 encoder responsible for the strip-like region adjacent to the lower side. The image after the lowest MB filter is transferred to the H.264 encoder 23. H. The H.264 encoder 23 receives the lowermost MB-filtered image of the strip-like region adjacent on the upper side, and stores the image in the memory. The H.264 encoder 23 handles the lowermost MB filtered image of the strip-shaped area. The data is transferred to the H.264 encoder 24. Similarly, H.M. H.264 encoder 24 is an H.264 encoder. The H.264 encoder 23 receives the image after the lowermost MB filter of the strip-like region adjacent on the upper side and stores it in the memory. The H.264 encoder 24 handles the lowermost MB filtered image of the strip-shaped area. It is transferred to the H.264 encoder 25. Similarly, H.M. H.264 encoder 25 is an H.264 encoder. From the H.264 encoder 24, the image after the lowest MB filter of the strip-like region adjacent on the upper side is received and stored in the memory. H. The image after the lowermost MB filter accumulated in the memory in the H.264 encoders 23 to 25 is used for loop filter processing to the macroblock at the uppermost end of each strip-shaped region by processing described later.

  H. of FIG. The configuration of the H.264 encoders 22 to 25 is the same as that of the conventional H.264 shown in FIG. The configuration of the H.264 encoder is partially different. H. according to an embodiment of the present invention. A configuration example of each of the H.264 encoders 22 to 25 is shown in FIG.

  Each H. H.264 that encodes strip-shaped areas in accordance with the H.264 standard. Each of the H.264 encoders 22 to 25 uses a strip-shaped area as an input image to be processed. The input strip-shaped area is input to the macroblock dividing unit 40. The macroblock dividing unit 40 divides the strip-like area into encoding processing units called macroblocks, and outputs them to the subtracter 41 as processing target macroblocks.

  At the same time as the input of the processing target macroblock, the inter prediction image macroblock is input to the subtractor 41 as the prediction macroblock when the intra / inter selection unit 42 selects inter. When the intra is selected, An intra prediction macroblock is input as a prediction macroblock. The subtractor 41 subtracts them in units of pixels in the macroblock and outputs the result to the integer DCT unit 43.

  The integer DCT unit 43 performs DCT conversion on the output of the subtracter 41 and outputs a DCT conversion coefficient to the quantization unit 44. The quantization unit 44 performs a quantization process at the designated quantization step, and outputs a quantization coefficient to the information source encoding unit 45. The information source coding unit 45 performs information source coding processing using a context adaptive variable length code called CAVLC or a context adaptive arithmetic code called CABAC, and outputs divided coded data.

  The quantization coefficient output from the quantization unit 44 is also input to the inverse quantization unit 46, where the inverse quantization unit 46 performs an inverse quantization process, and outputs an inverse-quantized DCT transform coefficient. The inversely quantized DCT transform coefficients are input to the inverse integer DCT unit 47, subjected to inverse integer DCT processing, and the converted macroblock is output to the adder 48. The adder 48 also receives the prediction macroblock, performs addition processing for each pixel, and outputs the local decode macroblock to the pre-filter frame memory 49. In the pre-filter frame memory 49, macroblocks are reconstructed in the raster scan order and recorded as a pre-filter strip area local decoded image.

  The pre-filter strip region local decoded image is input to the loop filter unit 50 when the loop filter is ON. The loop filter unit 50 removes block noise at the macroblock boundary from the input image, outputs it to the post-filter frame memory 51 as a post-filter strip region local decoded image, and accumulates it in the post-filter frame memory 51.

  In the loop filter unit 50, when the loop filter process is performed on the uppermost macroblock of each strip-shaped region, the bottom MB filter that is an image after the loop filter of the bottommost macroblock of the strip-shaped region directly above The subsequent image (input) is input to the loop filter unit 50 via the upper adjacent strip-shaped area bottom MB memory 56 from another encoder in charge of the adjacent strip-shaped area directly above.

  With this processing, when performing the loop filter processing of each macroblock at the uppermost end of each strip-shaped area, the loop filter unit 50 can acquire the image after the filter processing of the macroblock on the upper and left sides of the macroblock. Therefore, correct loop filter processing conforming to the standard becomes possible. Also, for loop filter processing of other encoders in charge of the strip area immediately below, an encoder in charge of the strip area adjacent to the lower MB filtered image (output) after the loop filter. Forward to.

  As an operation of turning off the loop filter, the pre-filter strip region local decoded image and the post-filter strip region local decoded image are the same, and the image stored in the pre-filter frame memory 49 is used as it is. It is transferred to the memory 51 and stored.

  Similar to the configuration shown in FIG. H.264 encoders 22 to 25 respectively obtain post-filter strip-like local decoded images owned by adjacent encoders by transferring the post-filter strip-like region local decoded images to each other. The filtered local decoded image of the entire input frame may be generated by combining the strip-shaped area local decoded images and recorded in the post-filter frame memory 51. By recording the locally decoded images of the upper and lower adjacent strip-shaped areas together in the post-filter frame memory 51, it becomes possible to extend the search range for motion prediction / motion compensation processing described later to outside the strip-shaped area. , Coding efficiency is improved.

  The motion prediction unit 52 receives the processing target macroblock and the filtered local decoded image, and performs motion estimation processing. The motion prediction unit 52 outputs the estimated motion vector to the motion compensation unit 53. The motion compensation unit 53 extracts a macroblock whose position is indicated by the motion vector in the filtered local decoded image, and outputs the macroblock to the intra / inter selection unit 42 as an inter predicted image macroblock.

  On the other hand, the pre-filter strip area local decoded image is input from the pre-filter frame memory 49 to the intra prediction mode evaluation unit 54 and the intra prediction value generation unit 55. H. In the H.264 standard, intra prediction is performed on the local decoded image before the processing in the loop filter unit 50 is performed. The intra prediction mode evaluation unit 54 selects an appropriate prediction mode from various selectable intra prediction modes, and outputs an intra prediction mode code to the intra prediction value generation unit 55. The intra prediction value generation unit 55 generates an intra prediction image macroblock from the pre-filter strip region local decoded image and the intra prediction mode code, and outputs the intra prediction image macroblock to the intra / inter selection unit 42.

  In FIG. 3, the pre-filter strip region local decoded image input to the loop filter unit 50 is input to the macroblock dividing unit 1. The macroblock division unit 1 divides the strip area into encoding processing units called macroblocks. However, unlike the corresponding circuit block (macroblock dividing unit 301) of the conventional configuration, the divided macroblocks are output to the vertical edge horizontal filter unit 2 as the processing target macroblocks in the vertical raster scan order. An example of processing in the macro raster vertical raster scan order is shown in FIG.

  At this time, in the loop filter unit 50 in the method shown in FIG. 3, the macroblock position calculation unit 15 calculates the positions of the macroblocks to be processed in the vertical raster scan order shown in FIG. As a result, the loop filter unit 50 performs the loop filter process in the vertical raster scan order, but the macroblock encoding process is similar to the configuration shown in FIG. Note that the horizontal raster scan order is used in accordance with the H.264 standard.

  The vertical edge horizontal filter unit 2 sequentially performs a filtering process in the horizontal direction on every vertical edge in every four pixels in the macroblock. FIG. 5 shows a processing outline of the vertical edge horizontal filter unit 2. Unlike the corresponding circuit block (with respect to the vertical edge horizontal filter unit 302) of the conventional configuration, the horizontal filter processing is not performed on the vertical edge a between the macroblock adjacent to the left of the processing target macroblock. Please keep in mind.

  The vertical edge horizontal filter unit b3 performs a filter process of 8 horizontal pixels × vertical 1 pixel with each pixel constituting the vertical edge b as the center. The pixel value obtained as a result of the filtering process on the vertical edge b is subjected to the filtering process on the vertical edge c in the horizontal edge filter unit c4. Similarly, as shown in FIG. 6, the horizontal filter processing for the vertical edge d is performed in the vertical edge horizontal filter section d5 on the horizontal filter processing result for the vertical edge c.

  The anti-horizontal edge vertical filter unit 6 has the same configuration as the anti-horizontal edge vertical filter unit 307 in the conventional configuration. That is, the horizontal edge vertical filter unit 6 sequentially performs filtering in the vertical direction on every horizontal edge in every four pixels in the macro block as shown in the right side of FIG.

  In the horizontal edge vertical filter unit e7, a filter process of horizontal 1 pixel × vertical 8 pixels is performed around each pixel constituting the horizontal edge e. This processing is performed using both the pixel value of the macroblock and the pixel value of the macroblock that has been subjected to the upper neighboring filter processing as input signal values. At this time, the image after the loop filter of the macro block adjacent above the macro block is supplied from the post-filter macro block memory 13 as the macro block after the upper filter processing. If the macro block to be filtered is at the uppermost end of the strip-shaped area, the strip-shaped area adjacent to the top immediately after being input to the post-filter macro-block memory 13 by the upper adjacent strip-shaped area bottom MB memory 56. Similarly, the filtered image of the lowermost macroblock of the region is input from the post-filter macroblock memory 13 to the horizontal edge vertical filter unit e7 as a macroblock that has been subjected to the upper adjacent filter processing.

  Similarly, as shown in FIG. 6, the vertical filter processing to the horizontal edge f is performed in the horizontal edge vertical filter unit f8 on the pixel value resulting from the filtering processing to the horizontal edge e. The horizontal edge vertical filter unit g9 performs a filtering process on the horizontal edge g on the vertical filter processing result on the horizontal edge f, and the output result on the horizontal edge h on the horizontal edge vertical filter unit h10. Is subjected to vertical filtering. The obtained mid-filter image (macroblock) is output to the vertical edge horizontal filter unit 11.

  In the vertical edge horizontal filter unit A12 in the horizontal edge horizontal filter unit 11, as shown in FIGS. 5 and 6, the vertical edge that is the vertical edge between the macroblock on the right side of the macroblock to be processed A filter process is performed on A in the horizontal direction. The post-filter macroblock obtained by performing the filtering process with respect to the vertical edge horizontal filter unit A12 is input to the post-filter macroblock memory 13, and the vertical raster scan for each macroblock is performed in the same manner as in the macroblock dividing unit 1. Reconstruction is performed in order, and it is recorded as a filtered strip-shaped area local decoded image.

  The filtered image of the lowermost MB in the strip-like area immediately above and inputted from the upper adjacent strip-like area lowermost MB memory 56 is inputted to the post-filter macroblock memory 13 and the uppermost of the strip-like area. This is used for loop filter processing to the MB horizontal edge e.

  Consider parallel coding processing using the moving image parallel coding processing configuration shown in FIGS. 1 and 2 and the loop filter processing configuration according to the method shown in FIG. H. Each loop filter unit 50 according to the present system in the H.264 encoders 22 to 25 is arranged in the vertical raster scan order from the upper left macro block to the lower right macro block as shown in FIG. Filter processing is performed.

  H. Loop filter between strip-shaped areas. In order to perform the filtering process in conformity with the H.264 standard, it is necessary that the upper adjacent macroblock has already been subjected to the filtering process when performing the vertical filter on the horizontal edge e. Therefore, as shown in FIG. 7, the filter processing is performed by shifting the row of macroblocks being processed in parallel between the strip-like regions. That is, from the left of the (X + 1) th strip-shaped divided region from the top after the filter processing in the vertical raster scan order of the rectangular small region group of the Yth column from the left of the X-th strip-shaped divided region from the top is completed. The filtering process in the vertical raster scan order of the rectangular small area in the Yth column is started.

By continuously performing the above operations in parallel, the following parallel filter processing operation is performed to realize moving image parallel coding.
(1) Start / complete vertical raster scan order filter processing for the first row of macroblocks in strip area 1
(2) Start and complete processing in parallel for the second row of strip-shaped area 1 and the macroblock group in the first row of strip-shaped area 2
(3) Processing starts and completes in parallel for the third row of strip-shaped area 1, the second row of strip-shaped area 2, and the first block of strip-shaped area 3
(4) Process the fourth row of strip-shaped area 1, the third row of strip-shaped area 2, the second row of strip-shaped area 3, and the macroblock group of the first row of strip-shaped area 4 in parallel. Start / Complete
(5) Thereafter, the B-th row of the strip-like region 1, the (B-1) -th row of the strip-like region 2, the (B-2) -th row of the strip-like region 3, and the (B -3) Processing is performed in parallel on the macroblock group in the column.

  That is, if this is generalized and explained, the filter processing of the C-th rectangular small area from the top of the rectangular small area group in the B-th column from the left of the A-th strip-shaped divided area from the top is performed. In parallel, another encoder performs a filtering process on the Cth rectangular small area from the top of the (A + B−W) th rectangular small area group from the left of the Wth strip-shaped divided area from the top, Realizes moving picture parallel coding.

  In addition, a loop filter between the strip-like regions is used. In order to perform filtering in conformity with the H.264 standard, the upper adjacent macroblock has already been filtered for the vertical filtering on the horizontal edge e that is the boundary of each strip-shaped region, and the upper adjacent The filtered image of the macroblock needs to be obtained. For this reason, before performing the vertical filter processing to the horizontal edge e to the uppermost macroblock of each strip-shaped region, after the filter processing of the lowermost macroblock of the upper adjacent strip-shaped region is completed, the filtered image is The encoder in charge of the upper adjacent strip-shaped area is transferred to and received from the encoder, stored in the memory, and used for filtering the uppermost macroblock of the strip-shaped area.

  As shown in FIG. Each of the H.264 encoders 22 to 25 can be configured as an LSI chip as an LSI chip. Each can also be realized by a computer and a software program.

  FIG. 8 is a flowchart of the filtering process in the moving picture coding method according to the embodiment of the present invention. In FIG. 8, the flow of the loop filter process for the Ath strip-shaped region from the top will be described. Since the flow of processing related to other related encoding is the same as that of the prior art, description using a flowchart is omitted.

  In step S1, the macroblock position calculation unit 15 initializes the position of the processing target macroblock. Here, when processing the macroblock to be processed as a macroblock at the position of the Bth column from the left and the Cth row from the top in the Ath strip-shaped region, it is initialized to B = 0 and C = 0.

  In step S2, the macroblock dividing unit 1 extracts the macroblocks in the B-th column and the C-th row as processing target macroblocks from the pre-filter strip-shaped region local decoded image. Thereafter, in response to an instruction from the macroblock position calculation unit 15, the processing target macroblocks are extracted in the order of the vertical raster scan.

  In step S3, the vertical edge horizontal filter unit 2 performs a filter process of 8 horizontal pixels × vertical 1 pixel centering on each pixel constituting the vertical edge b. Further, the filtering process for the vertical edge c is performed on the pixel value obtained as a result of the filtering process for the vertical edge b. Similarly, the horizontal filter processing for the vertical edge d is performed on the horizontal filter processing result for the vertical edge c.

  In step S4, the horizontal edge vertical filter unit 6 sequentially performs a filtering process in the vertical direction on every horizontal edge in the macroblock. First, a filter process of 1 horizontal pixel × 8 vertical pixels is performed around each pixel constituting the horizontal edge e. Further, the vertical filter processing for the horizontal edge f is performed on the pixel value obtained as a result of the filtering processing for the horizontal edge e. Similarly, a filtering process for the horizontal edge g and a vertical filtering process for the horizontal edge h are performed.

  In step S5, the vertical edge horizontal filter unit 11 performs horizontal filtering on the vertical edge A, which is an adjacent boundary between the processing target macroblock and the right macroblock, as the last filtering step. Do.

  In step S6, the filtered macroblock is recorded and stored at the position of the macroblock in the Bth column and the Cth row in the filtered macroblock memory 13.

  Next, in step S7, it is determined whether or not the filter processing has been completed for all macroblocks in the Bth column from the left in the Ath strip-shaped region. If not, the process proceeds to step S8. If completed, the process proceeds to step S9.

  In step S8, the macroblock position calculation unit 15 adds 1 to C and sets the next lower macroblock as the processing target macroblock. Thereafter, the processing of steps S2 to S6 is repeated in the same manner for this processing target macroblock.

  In step S9, it is determined whether or not the filter processing has been completed for all the macroblocks in the A-th strip-shaped area from the top, and if not, the process proceeds to step S10. If completed, the loop filter process for the A-th strip-shaped region from the top is finished.

  In step S10, the macroblock position calculation unit 15 initializes C to 0, adds 1 to B, and sets the uppermost macroblock in the right adjacent macroblock column of the current macroblock as the processing target macroblock. To do.

  In step S11, it is determined whether or not the filtering process has been completed for all the B + 1th macroblocks in the strip area immediately above (A-1), and if not, the process is awaited. When the filtering process for all the B + 1th macroblocks in the strip area immediately above (A-1) is completed, the processes after step S2 are similarly repeated.

  In the description of the above embodiment, H.C. The application example of the present invention to the parallel coding conforming to the H.264 standard has been described. It is not limited to H.264 encoding. Even in the case of a new standard that is an improvement of the H.264 standard, for the application of the present invention, H.264 standard. It goes without saying that the present invention can be applied in exactly the same manner as long as it has a technical content similar to that of the H.264 standard.

  The above moving image encoding processing can be realized by a computer and a software program, and the program can be provided by being recorded on a computer-readable recording medium or provided through a network. .

  In MPEG-2, there has been proposed a method for realizing HD (high definition) CODEC using a plurality of SD (standard television) scale CODEC-LSIs (see Non-Patent Document 2). The above technique is used as it is. By extending to H.264, HD compatible H.264 264-CODEC is feasible, but simple H.264. For application to H.264, H.264. The loop filter processing defined by the standard that is the feature of H.264 cannot be realized. The present invention relates to H.264. H.264 compatible with loop filter processing conforming to the H.264 standard. 264-HD-CODEC, a number of small H.264 It can be realized by parallel processing using 264-LSI.

It is a figure which shows the example of a whole structure of the moving image parallel encoding apparatus containing this invention. H. according to an embodiment of the present invention. 2 is a diagram illustrating a configuration example of an H.264 encoder. FIG. It is a loop filter block diagram concerning the Example of this invention. It is a figure explaining the loop filter vertical raster scan system in the Example of this invention. It is a figure which shows the process outline | summary of the versus vertical edge horizontal filter part in the Example of this invention. It is a figure which shows the horizontal loop filter process in the Example of this invention, and a vertical loop filter process order. It is a figure which shows the parallel filter process timing which employ | adopted the vertical raster scan forward loop filter in the Example of this invention. It is a flowchart of the filter process in the moving image encoding method which concerns on the Example of this invention. It is a figure which shows the parallel encoding process structure based on the conventional MPEG-2 standard. It is a figure which shows the strip-shaped area | region division example of an input frame. It is a figure which shows the example of a conventional MPEG-2 encoder structure. Conventional H.264. 2 is a diagram illustrating a configuration example of an H.264 encoder. FIG. General H.P. It is a figure which shows a H.264 loop filter process order. General H.P. It is a figure which shows a H.264 loop filter structural example. Conventional H.264. It is a figure explaining the H.264 horizontal loop filter process and the vertical loop filter process outline. Conventional H.264. It is a figure which shows a H.264 horizontal loop filter process and a vertical loop filter process order. Conventional H.264. It is a figure explaining the problem in a H.264 parallel encoding process.

Explanation of symbols

1 Macroblock division unit 2,11 pair vertical edge horizontal filter unit 3-5,12 pair vertical edge horizontal filter unit bd, A
6 to horizontal edge vertical filter part 7 to 10 to horizontal edge vertical filter part e to h
13 Macro-block memory after filter 15 Macro-block position calculation unit 20 Video parallel encoding device 21 Screen division processing unit 22-25 H.264 encoder 26 Stream concatenation unit 40 Macroblock division unit 41 Subtractor 42 Intra-inter selection unit 43 Integer DCT unit 44 Quantization unit 45 Information source encoding unit 46 Inverse quantization unit 47 Inverse integer DCT unit 48 Adder 49 Filter Pre-frame memory 50 Loop filter unit 51 Post-filter frame memory 52 Motion prediction unit 53 Motion compensation unit 54 Intra prediction mode evaluation unit 55 Intra prediction value generation unit 56 Upper adjacent strip-shaped region bottom MB memory

Claims (15)

Each frame of the input moving image is divided into regions in the horizontal direction to generate a plurality of strip-shaped divided regions, and a plurality of encoders respectively divide each strip-shaped divided region into rectangular small regions in a grid pattern. A moving picture coding method executed by each of the encoders for coding the strip-like divided areas used in a moving picture parallel coding apparatus for coding rectangular small areas in a parallel divided area in parallel. ,
For rectangular small areas obtained by dividing the strip-shaped divided area into a grid, each rectangle from the upper-left rectangular small area in the strip-shaped divided area toward the lower-right rectangular small area in the strip-shaped divided area Encoding for each small region in the order of horizontal raster scan;
For the rectangular small area obtained by dividing the strip-shaped divided area of the local decoded image obtained by decoding the encoded data of the encoded strip-shaped divided area into a lattice shape, the upper left rectangular small area in the strip-shaped divided area Performing a filtering process in the order of vertical raster scan for each rectangular small area from the rectangular divided area toward the lower right rectangular small area in the strip-shaped divided area;
When there is an encoder in charge of encoding the strip-like divided area adjacent to the upper side of the strip-like divided area, it is adjacent to at least the uppermost rectangular small area of the strip-like divided area from the encoder. Receiving a filtered image of a rectangular small area to be recorded in a memory;
When there is an encoder responsible for encoding the strip-like divided area adjacent to the lower side of the strip-like divided area, at least the lowest rectangular small part of the strip-like divided area is directed toward the encoder. Transferring the filtered image of the region,
In the step of performing the filtering process, when performing the filtering process on the uppermost rectangular small area of the strip-shaped divided area, the filtering process is performed using the image after the filtering process recorded in the memory,
In the encoding step, the inter-frame prediction encoding of the subsequent frame is performed using the local decoded image after the filtering process as a reference image. In the moving image encoding method according to claim 1,
In the step of performing the filtering process, in the filtering process of one rectangular small area in the strip-shaped divided area, a horizontal filtering process for a vertical edge for pixels in the rectangular small area, and a horizontal direction After performing vertical filtering on the edges of the rectangle, horizontal filtering is performed on the vertical edges for the border pixels between the rectangular small area and the rectangular small area adjacent to the right of the rectangular small area. A moving image encoding method characterized by comprising: In the moving image encoding method according to claim 1 or 2,
After the encoding step in the horizontal raster scan order of all the rectangular small areas obtained by dividing the strip-shaped divided areas into a grid pattern by the encoding step,
The moving image encoding method, wherein the filtering process is started by performing the filtering process. Each frame of the input moving image is divided into regions in the horizontal direction to generate a plurality of strip-shaped divided regions, and a plurality of encoders respectively divide each strip-shaped divided region into rectangular small regions in a grid pattern. An encoder for encoding the strip-shaped divided region used in a moving image parallel coding device for encoding rectangular small regions in a rectangular divided region in parallel;
For rectangular small areas obtained by dividing the strip-shaped divided area into a grid, each rectangle from the upper left rectangular small area in the strip-shaped divided area toward the lower right rectangular small area in the strip-shaped divided area Means for encoding each small area in the order of horizontal raster scan;
For the rectangular small area obtained by dividing the strip-shaped divided area of the local decoded image obtained by decoding the encoded data of the encoded strip-shaped divided area into a lattice shape, the upper left rectangular small area in the strip-shaped divided area Means for performing a filtering process in the order of vertical raster scan for each rectangular small area from the rectangular divided area toward the lower right rectangular small area;
When there is an encoder in charge of encoding the strip-like divided area adjacent to the upper side of the strip-like divided area, it is adjacent to at least the uppermost rectangular small area of the strip-like divided area from the encoder. Means for receiving a filtered image of a rectangular small area and recording it in a memory;
When there is an encoder responsible for encoding the strip-like divided area adjacent to the lower side of the strip-like divided area, at least the lowest rectangular small part of the strip-like divided area is directed toward the encoder. Means for transferring the image after region filtering,
The means for performing the filtering process performs the filtering process using the image after filtering recorded in the memory when performing the filtering process on the uppermost rectangular small area of the strip-shaped divided area,
The moving image parallel coding encoder, wherein the coding means performs inter-frame predictive coding of a subsequent frame using the filtered local decoded image as a reference image. The encoder for parallel encoding of moving images according to claim 4,
The means for performing the filtering process includes: a filtering process for one rectangular small area in the strip-shaped divided area; a horizontal filtering process for a vertical edge for a pixel in the rectangular small area; After performing vertical filtering on the edges of the rectangle, horizontal filtering is performed on the vertical edges for the border pixels between the rectangular small area and the rectangular small area adjacent to the right of the rectangular small area. An encoder for parallel video encoding, characterized in that: Each frame of the input moving image is divided into regions in the horizontal direction to generate a plurality of strip-shaped divided regions, and a plurality of encoders respectively divide each strip-shaped divided region into rectangular small regions in a grid pattern. A video parallel encoding method for encoding rectangular small regions in a segmented region in parallel,
For each rectangular small area obtained by dividing each strip-shaped divided area in a grid pattern by each of the encoders, a rectangular shape located at the lower right in the strip-shaped divided area from the rectangular small area at the upper left in the strip-shaped divided area. Encoding in a horizontal raster scan order for each rectangular subregion towards the subregion;
For each rectangular small region obtained by dividing each strip-shaped divided region of the local decoded image obtained by decoding the encoded data of the encoded strip-shaped divided region into a lattice shape, each encoder has the strip-shaped divided region. Performing a filtering process in the order of vertical raster scan for each rectangular small region from the rectangular small region at the upper left to the rectangular small region at the lower right within the strip-shaped divided region in
When there is an encoder in charge of encoding the strip-like divided area adjacent to the upper side of the strip-like divided area, each encoder has at least the most of the strip-like divided area. Receiving a filtered image of a rectangular small area adjacent to the upper rectangular small area and recording it in a memory;
When each of the encoders has an encoder responsible for encoding the adjacent strip-shaped divided region below the strip-shaped divided region, at least the strip-shaped divided portion is directed toward the encoder. Transferring the filtered image of the rectangular small area at the bottom of the area,
In the step of performing the filtering process, when performing the filtering process on the uppermost rectangular small area of the strip-shaped divided area, the filtering process is performed using the image after the filtering process recorded in the memory,
In the encoding step, the inter-frame predictive encoding of the subsequent frame is performed using the local decoded image after the filtering process as a reference image. In the moving image parallel encoding method according to claim 6,
In each of the encoders performing the filtering process, in the filtering process for one rectangular small area in the strip-shaped divided area, a horizontal direction with respect to a vertical edge for pixels in the rectangular small area is selected. After the filtering process and the vertical filtering process for the horizontal edge, the vertical edge for the adjacent boundary pixel between the rectangular small area and the rectangular small area adjacent to the right of the rectangular small area is processed. A moving image parallel coding method characterized by performing horizontal filtering. In the moving image parallel encoding method according to claim 6 or 7,
Each encoder is
After the encoding step in the horizontal raster scan order of all the rectangular small areas obtained by dividing the strip-shaped divided areas into a grid pattern by the encoding step,
The moving image parallel encoding method, wherein the filtering process by the step of performing the filtering process is started. In the moving image parallel encoding method according to claim 6, claim 7 or claim 8,
In parallel with one of the encoders performing the filtering process of the C-th rectangular small region from the top of the rectangular small region group of the B-th column from the left of the A-th strip-shaped divided region from the top, The another encoder performs a filtering process on the C-th rectangular small region from the top of the (A + B−W) -th rectangular small region group from the left of the W-th strip-shaped divided region from the top. A moving image parallel coding method. Each frame of the input moving image is divided into regions in the horizontal direction to generate a plurality of strip-shaped divided regions, and a plurality of encoders respectively divide each strip-shaped divided region into rectangular small regions in a grid pattern. A video parallel encoding device that encodes rectangular small regions in a segmented region in parallel,
Each encoder is
For rectangular small areas obtained by dividing the strip-shaped divided area into a grid, each rectangle from the upper-left rectangular small area in the strip-shaped divided area toward the lower-right rectangular small area in the strip-shaped divided area Means for encoding each small area in the order of horizontal raster scan;
For the rectangular small area obtained by dividing the strip-shaped divided area of the local decoded image obtained by decoding the encoded data of the encoded strip-shaped divided area into a lattice shape, the upper left rectangular small area in the strip-shaped divided area Means for performing a filtering process in the order of vertical raster scan for each rectangular small area from the rectangular divided area toward the lower right rectangular small area;
When there is an encoder in charge of encoding the strip-like divided area adjacent to the upper side of the strip-like divided area, it is adjacent to at least the uppermost rectangular small area of the strip-like divided area from the encoder. Means for receiving a filtered image of a rectangular small area to be recorded in a memory;
When there is an encoder responsible for encoding the strip-like divided area adjacent to the lower side of the strip-like divided area, at least the lowest rectangular small part of the strip-like divided area is directed toward the encoder. Means for transferring the image after region filtering,
The means for performing the filtering process performs the filtering process using the image after filtering recorded in the memory when performing the filtering process on the uppermost rectangular small area of the strip-shaped divided area,
The moving picture parallel coding apparatus characterized in that the coding means performs inter-frame predictive coding of subsequent frames using the filtered local decoded image as a reference image. In the moving image parallel encoding device according to claim 10,
The means for performing the filtering process of each encoder includes: a filtering process for one rectangular small area in the strip-shaped divided area; and a horizontal direction with respect to a vertical edge for pixels in the rectangular small area. After the filtering process and the vertical filtering process for the horizontal edge, the vertical edge for the adjacent boundary pixel between the rectangular small area and the rectangular small area adjacent to the right of the rectangular small area is processed. A moving image parallel coding apparatus characterized by performing horizontal filtering.   A moving picture coding program for causing a computer to execute the moving picture coding method according to any one of claims 1 to 3.   A computer-readable recording medium having recorded thereon a moving image encoding program for causing a computer to execute the moving image encoding method according to any one of claims 1 to 3.   A moving picture parallel coding program for causing a computer to execute the moving picture parallel coding method according to any one of claims 6 to 9.   A computer-readable recording medium on which a moving image parallel coding program for causing a computer to execute the moving image parallel coding method according to any one of claims 6 to 9 is recorded.

Images