[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

WO2016155070A1 - 一种面向多纹理多深度视频的相邻视差矢量获取方法 - Google Patents

一种面向多纹理多深度视频的相邻视差矢量获取方法 Download PDF

Info

Publication number
WO2016155070A1
WO2016155070A1 PCT/CN2015/077944 CN2015077944W WO2016155070A1 WO 2016155070 A1 WO2016155070 A1 WO 2016155070A1 CN 2015077944 W CN2015077944 W CN 2015077944W WO 2016155070 A1 WO2016155070 A1 WO 2016155070A1
Authority
WO
WIPO (PCT)
Prior art keywords
disparity vector
disparity
coding unit
adjacent
group
Prior art date
Application number
PCT/CN2015/077944
Other languages
English (en)
French (fr)
Inventor
贾克斌
周作成
Original Assignee
北京工业大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京工业大学 filed Critical 北京工业大学
Priority to US15/316,009 priority Critical patent/US9883200B2/en
Publication of WO2016155070A1 publication Critical patent/WO2016155070A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H04N19/513—Processing of motion vectors
    • H04N19/517—Processing of motion vectors by encoding
    • H04N19/52—Processing of motion vectors by encoding by predictive encoding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
    • H04N19/146—Data rate or code amount at the encoder output
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/597—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding specially adapted for multi-view video sequence encoding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
    • H04N19/17—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object
    • H04N19/172—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object the region being a picture, frame or field

Definitions

  • the present invention relates to a video coding technology based on 3D-HEVC, and in particular to a method for acquiring adjacent disparity vectors in auxiliary view coding in multi-texture multi-depth video coding.
  • the multi-depth multi-texture video used in the 3D-HEVC standard is currently the most effective 3D video representation: multi-depth multi-texture video is performed by multiple cameras (usually shooting 3 textures and 3 depths) from different angles to the same scene.
  • the multi-texture video with multiple depth information is captured, and the multi-texture video with multi-depth information can describe the 3D scene information in a more detailed and comprehensive manner, which is beneficial to the 3D video terminal in the larger viewing angle range and adopts the depth-based virtual viewpoint.
  • the rendering technique produces a high-quality virtual view of any angle of view, while providing both binocular parallax and motion parallax, providing an immersive viewing experience for the user.
  • the horizontal direction is the time direction and the vertical direction is the viewpoint direction.
  • Use a layered B-frame structure in the time direction to eliminate redundancy in the time direction, between viewpoints Eliminate redundant information between pilots using an I-P-P architecture.
  • the primary view can only use the coded frame in the current view as the reference frame, and the auxiliary view can use the coded frame of the primary view as the reference frame in addition to the frame that has been encoded in the present view.
  • a depth map containing the corresponding 8-bit representation is included.
  • the main view texture map is first encoded, then the main view depth map is encoded, and then the auxiliary view is subjected to texture map and depth map coding. Because the depth map requires the encoding information of the texture map to be encoded, and the auxiliary viewpoint encoding needs to be encoded by using the encoding information of the main viewpoint, which is called the texture encoding priority encoding sequence, as shown in FIG. 2 .
  • the acquisition of disparity vectors is a key technique in multi-texture and multi-depth 3D-HEVC video coding technology, and is widely used in inter-view motion compensation prediction and inter-view residual prediction.
  • the disparity vector represents the difference between different video frames at the same time.
  • the disparity vector in the main view can perform motion compensation on the prediction unit of the auxiliary view.
  • the texture map coding priority coding sequence when the prediction unit in the auxiliary view is encoded, the disparity vector used by it cannot be calculated from the corresponding depth map because its corresponding depth map has not yet been input into the coding unit.
  • the traditional disparity vector acquisition method is obtained by block estimation and block matching, and relevant information is also needed at the decoding end for decoding operations. If this information is transmitted in the code stream, additional transmission bits will be generated.
  • a technique for estimating a depth map from encoded texture map information has been introduced in the existing 3D-HEVC.
  • the disparity information between the primary viewpoint and the secondary viewpoint is often converted into depth map information, and the calculated depth map information can be used to convert the primary viewpoint and other auxiliary viewpoint depth map information.
  • the maximum depth value in the depth map estimated in this process will be converted into a disparity vector, which is called a depth map based disparity vector conversion.
  • a simplified disparity vector acquisition algorithm is introduced to reduce computational complexity 3D-HEVC, which is called a disparity vector acquisition method based on neighboring blocks.
  • the disparity vector acquisition method based on the neighboring block searches the candidate space and the time coded block position according to a preset order, and determines whether the disparity vector information is included to obtain the disparity vector of the current block, and the spatial and temporal coding block positions are as shown in FIG. 3 and FIG. 4 . Shown. If the searched prediction unit uses parallax compensation prediction or disparity motion vector compensation prediction technology, it indicates that the prediction unit includes disparity information, and the first searched disparity vector is used for inter-view motion compensation prediction and inter-view disability.
  • the process of poor prediction is: first search for the positions CRT and BR in the time reference frame, then search for the spatial reference frame positions A1, B1, B0, A0 and B2, and finally search for the motion vector compensation situation in the spatial reference frame position.
  • the disparity vector acquisition method based on the neighboring block saves at least 8% or more time than the disparity vector conversion method based on the depth map.
  • the 3D-HEVC further improves the acquired adjacent disparity vector by using the depth map-based neighbor block disparity vector acquisition method.
  • the depth map-based neighboring block disparity vector acquisition method uses the encoded main view depth map to correct the initially acquired disparity vector. After the disparity vector of the original neighboring block is obtained, the maximum depth value of the depth map of the relevant main view is used to correct and obtain the final disparity vector.
  • the existing improved algorithm based on neighboring block-based disparity vector acquisition puts the research focus on the reduced search candidate position to reduce the number of searches.
  • the main problem is that the first searched disparity vector is regarded as the final disparity vector, the search stops, and the remaining search positions may still contain the disparity vector that can be utilized or even better before it is searched. It was stopped by the entire acquisition process. Therefore, the present invention is based on the 3D-HEVC standard, and proposes a method for changing the first searched disparity vector to be regarded as the criterion of the final disparity vector, by deleting the candidate space and the time coding unit position adjacent to the current coding unit. The least searched position, the adjacent candidate space and the time coding unit position are grouped, and the searched disparity vector is combined into the final disparity vector according to the ratio of the adoption rate, which can maintain the original fast algorithm. Improve coding quality under the premise of efficiency.
  • the object of the present invention is to propose a method for changing the first searched disparity vector in the existing block-based disparity vector acquisition method, which is regarded as the final disparity vector, based on the 3D-HEVC standard. Deleting the candidate space adjacent to the current coding unit and the least searched position of the time coding unit position, and grouping the adjacent candidate space and the time coding unit position, and combining the searched disparity vectors according to the ratio of the adoption rate to The final parallax vector method can improve the encoding quality while maintaining the efficiency of the original fast algorithm.
  • Step (1) computer initialization:
  • the texture map and the corresponding depth map quantization parameters are divided into Group 1 (40, 45), Group 2 (35, 42), Group 3 (30, 39), Group 4 (25, 34). ), where the number in front of the brackets represents the texture map quantization parameter, followed by the depth map quantization parameter, a total of 4 groups.
  • the 3D-HEVC international standard test video sequence in YUV format includes: Newspaper_CC (300 frames), GT fly (250 frames), Undo dancer (250 frames), Poznan_Hall2 (200 frames), Poznan_Street (250 frames), Kendo (300) Frame), Balloons (300 frames), each test sequence contains 3 texture sequences, and corresponding 3 depth sequences.
  • the three viewpoint sequences of the input encoder when encoding each test sequence are: 4-2-6, 5-9-1, 5-1-9, 6-7-5, 4-5-3, 3-1 -5, 3-1-5, in the case of encoding each viewpoint, the texture map encoding is completed and then the depth map encoding is performed.
  • the HTM encoding software running on the computer first reads the main frame of the first frame number 4 after being initialized according to the general test condition texture map and the corresponding depth map quantization parameter set (40, 45).
  • the sequence texture map is encoded, and then the main view depth map of the first frame number 4 is encoded, and after the first frame encoding of the main view is completed, the auxiliary view sequence texture map of the first frame number 2 is read and encoded.
  • the auxiliary view sequence depth map of the first frame number 2 is read and encoded
  • the auxiliary view sequence texture map of the first frame number 6 is read and encoded, and the auxiliary view number of the first frame number 6 is read after completion.
  • the depth map is encoded.
  • the first frame encoding of the three view sequences is completed, and 300 frames are sequentially read to complete the encoding of the Newspaper_CC sequence under the quantization parameter group (40, 45).
  • the HTM encoding software running on the computer first reads the first view sequence texture map of the first frame number 4 for encoding, and then reads.
  • the main view depth map of the first frame number 4 is encoded, and after the first frame encoding of the main view is completed, the auxiliary view sequence texture map of the first frame number 2 is read and encoded, and the first frame number is read after completion.
  • the auxiliary view sequence depth map of 2 is encoded, and the auxiliary view sequence texture map of the first frame number 6 is read and encoded, and the auxiliary view depth map of the first frame number 6 is read and encoded.
  • the first frame encoding of the three view sequences is completed, and 300 frames are sequentially read to complete the encoding of the Newspaper_CC sequence under the quantization parameter group (35, 42).
  • the encoding in the Newspaper_CC sequence under the quantization parameter sets (30, 39) and (25, 34) is then completed in sequence.
  • the HTM encoding software running on the computer first reads the first frame number 5
  • the main view sequence texture map is encoded, and then the main view depth map of the first frame number 5 is read and encoded.
  • the first frame is read.
  • the auxiliary view sequence texture map with the number 9 is encoded, and after reading, the auxiliary view sequence depth map of the first frame number 9 is encoded, and finally the auxiliary view sequence texture map with the first frame number 1 is read and encoded. Then, the auxiliary view depth map of the first frame number 1 is read and encoded.
  • the first frame encoding of the three view sequences is completed, and the encoding of the GT fly sequence under the quantization parameter group (40, 45) is sequentially read to 250 frames.
  • the HTM encoding software running on the computer first reads the main view sequence texture map with the first frame number 5, and then reads.
  • the main view depth map of the first frame number is 5, and after the first frame of the main view is encoded, the auxiliary view sequence texture image of the first frame number 9 is read and encoded, and the first frame number is read after completion.
  • the auxiliary view sequence depth map of the first frame number is 1 is encoded, and the auxiliary view depth map of the first frame number 1 is read and encoded.
  • the first frame encoding of the three view sequences is completed, and the encoding of the GT fly sequence under the quantization parameter group (35, 42) is sequentially read to 250 frames.
  • the encoding of the GT fly sequence under the quantization parameter sets (30, 39) and (25, 34) is completed in sequence;
  • Step (2) using the video encoding software HTM8.0 to select the first n frames (n) of the 3D-HEVC international standard test video sequence Newspaper_CC, GT fly, Undo dancer, Poznan_Hall2, Poznan_Street, Kendo, Balloons, respectively.
  • the main view and the auxiliary view are encoded for a natural number greater than or equal to 40 and less than or equal to 60.
  • the encoding of each view is performed by using texture map coding and then performing depth map coding.
  • the all-position search method is adopted for the candidate spatial and temporal coding unit positions adjacent to the current coding unit, and the candidate vector space of the current coding unit and the disparity vector adoption rate of the temporal coding unit position are simultaneously extracted.
  • the information includes: detection of adjacent candidate temporal coding unit positions CTR and BR and whether adjacent candidate spatial coding unit positions A1, B1, B0, A0, B2 contain disparity vectors or motion compensated prediction disparity vectors, and Performing a difference sum of squares (SSD, Sum of Squared Difference) on the disparity vector found by all adjacent candidate spatial and temporal coding unit positions of the current coding unit or the disparity reference frame pointed by the motion compensated prediction disparity vector information, and the difference is squared
  • SD Sum of Squared Difference
  • the HTM encoding software running on the computer first reads the assistance of the first frame when encoding the auxiliary viewpoint. Viewing the texture map, and then reading the first code block of the current texture map, searching for the corresponding coding unit positions CTR and BR in the first coding block time direction reference frame, detecting whether the disparity vector is included, and searching for the coding in the spatial direction reference frame. Whether the unit position A1, B1, B0, A0, B2 contains a disparity vector, and then searches for the coding unit in the spatial direction reference frame.
  • the position A1, B1, B0, A0, B2 contains a motion compensated prediction disparity vector, for example, in the first coding block, it is found that A0 and A1 both contain a disparity vector, and the disparity vector is respectively found by the A0 position and the A1 position disparity vector.
  • the corresponding coding unit in the reference frame respectively performs the square sum calculation of the difference. If the sum of the squares of the A1 position differences is the smallest, it is recorded as the A1 position once.
  • the depth map coding sequence is read in the first 40 frames to complete the sequence Newspaper_CC in the quantization parameter group (40, 45) The number of times the position is counted. Then, the sequence News_CC is encoded in the first n frames (n 40) under the quantization parameter group (35, 42), and the HTM encoding software running on the computer first reads the auxiliary view texture of the first frame when encoding the auxiliary viewpoint.
  • the first coding block of the current texture map is read, and the corresponding coding unit positions CTR and BR in the first coding block time direction reference frame are searched for whether or not the disparity vector is detected, and then the coding unit position in the spatial direction reference frame is searched.
  • A1, B1, B0, A0, B2 contain a disparity vector
  • both A0 and A1 contain disparity vectors, and then the corresponding coding units in the disparity reference frame are respectively calculated by the A0 position and the A1 position disparity vector to calculate the sum of squares of differences, if the sum of the squares of the A1 position differences is the smallest, then Recorded as A1 position once.
  • the all-coded block of the first frame is sequentially read to complete the auxiliary view texture map of the first frame, and the number of times is counted, and the depth map is encoded according to the sequence of the texture map, and the sequence of the first 40 frames is completed.
  • Newspaper_CC is in the quantization parameter group (35) , 42) The number of times the position is counted.
  • the statistics of the number of times in the Newspaper_CC sequence under the quantization parameter groups (30, 39) and (25, 34) are sequentially completed, and finally the number of times of each position is divided by the total number of times to obtain the statistics of the adoption rate of each position of the Newspaper_CC sequence.
  • the HTM encoding software running on the computer first reads the first frame when encoding the auxiliary view.
  • the auxiliary view texture map and then reading the first code block of the current texture map, searching for the corresponding coding unit positions CTR and BR in the first coding block time direction reference frame, detecting whether the disparity vector is included, and then searching for the spatial direction reference frame.
  • the median unit position A1, B1, B0, A0, B2 contains a disparity vector
  • the coding unit position A1, B1, B0, A0, B2 in the spatial direction reference frame contains a motion compensated prediction disparity vector
  • the corresponding coding units in the disparity reference frame are respectively found by the A0 position and the A1 position disparity vector to calculate the sum of the differences, if the A1 position difference is the sum of squares For the smallest, it is recorded as the A1 position once.
  • the all-coded block of the first frame is sequentially read to complete the auxiliary view texture map of the first frame, and the number of times is counted, and the sequence of the first 40 frames is read in accordance with the sequence of the depth map encoding.
  • the number of times the GT fly is in the position of the quantization parameter set (40, 45).
  • the sequence GT fly is encoded in the first n frames (n 40) under the quantization parameter group (35, 42), and the HTM encoding software running on the computer first reads the auxiliary view of the first frame when encoding the auxiliary viewpoint.
  • the texture map and then reading the first coding block of the current texture map, searching for the corresponding coding unit positions CTR and BR in the first coding block time direction reference frame, detecting whether the disparity vector is included, and searching for the coding unit in the spatial direction reference frame. Whether the position A1, B1, B0, A0, B2 contains a disparity vector, and then whether the coding unit position A1, B1, B0, A0, B2 in the spatial direction reference frame contains a motion compensated prediction disparity vector, for example, in the first coding If both blocks A0 and A1 are found to contain disparity vectors, the corresponding coding units in the disparity reference frame are respectively found by the A0 position and the A1 position disparity vector to calculate the sum of the squares of the differences.
  • the all-coded block of the first frame is sequentially read to complete the auxiliary view texture map of the first frame, and the number of times is counted, and the depth map coding is performed according to the texture map coding, and the first 40 frames complete sequence GT fly is used in the quantization parameter group ( The number of times the position of 35, 42) is counted. Then, the statistics of the number of times in the GT fly sequence under the quantization parameter groups (30, 39) and (25, 34) are sequentially completed. Finally, the number of times of each position is divided by the total number of times to obtain the statistics of the adoption rate of each position of the GT fly sequence. The statistical results of the adoption rate of each position are shown in Table 1 and Table 2;
  • Step (3) the coding units of the coded frames of all the auxiliary view points except the main view point in the multi-texture multi-depth video are grouped according to the candidate space and the time coding unit position adjacent to the current coding unit, and are used according to the adoption rate.
  • the proportional combination is the final disparity vector method for adjacent disparity vector acquisition, and the steps are as follows:
  • Step (3.1) using the information of the disparity vector adoption rate of the candidate space adjacent to the current coding unit and the position of the temporal coding unit obtained by the step (2), provides a basis for the group search: first, the adjacent candidate time and space coding unit position
  • the deletion operation is performed to delete the coding unit position with the smallest disparity vector adoption rate.
  • the statistical result of each position obtained in the Newspaper_CC sequence is: the coding unit position A1, B1, B0, A0, B2 in the spatial direction reference frame is 65.7%, 0.9%, 3.0%, 30.1%, 0.2. %, the corresponding coding unit position CTR and BR adoption rate in the time direction reference frame is 76.6%, 23.4%.
  • the coding unit position B2 in the last spatial direction reference frame is deleted, and the corresponding coding unit position BR in the time direction reference frame is deleted.
  • the adoption rate statistics of the last positions in the GT fly sequence are: the coding unit positions A1, B1, B0, A0, and B2 in the spatial direction reference frame are 66.7%, 2.1%, 6.0%, 24.2%, 1.0. %, the corresponding coding unit position CTR and BR adoption rate in the time direction reference frame is 78.5%, 21.5%.
  • the coding unit position B2 in the last spatial direction reference frame is deleted, and the corresponding coding unit position BR in the time direction reference frame is deleted.
  • the rest of the test sequence is obtained according to the statistical results of the adoption rate.
  • the coding unit position B2 is deleted in the test frame, and the corresponding coding unit position BR in the time direction reference frame is deleted;
  • Step (3.2) grouping adjacent candidate space and time coding unit positions after deletion, and obtaining, by step (3.1), A1, B1, B0, and A0 of the reference search position including the spatial position and the CRT of the time reference position , as shown in Figure 7 and Figure 8.
  • the grouping operation is for coding units of adjacent positions, where B1 and B0 are adjacent, A1 and A0 are adjacent, then B1 and adjacent B0 are grouped, and A1 and adjacent A0 are divided into another group, for example: in Newspaper_CC
  • the coding unit position B2 in the spatial direction reference frame in the sequence is deleted, and the candidate spatial coding unit positions A1 and B1 coding unit positions and their adjacent A0 and B0 coding unit positions form a group.
  • the coding unit position B2 is deleted in the spatial direction reference frame, and the candidate spatial coding unit positions A1 and B1 coding unit positions and their adjacent A0 and B0 coding unit positions form a group.
  • the coding unit position B2 in the remaining sequence spatial direction reference frame is deleted, and the candidate spatial coding unit positions A1 and B1 coding unit positions and their adjacent A0 and B0 coding unit positions form a group.
  • the combination ratio of A1 and A0 and B1 and B0 after grouping is shown in Table 3;
  • Step (3.3) the current coding unit obtained by the step (3.2) performs a search for the adjacent disparity vector and a synthetic calculation operation of the final disparity vector for the adjacent candidate spatial and temporal coding unit search position and the group information, and the steps are as follows:
  • Step (3.3.1) set the acquisition disparity vector flag variable, and the disparity vector flag variable characterizes whether the disparity vector is acquired. If it has been acquired, it is 1 and if it is not acquired, it is 0 value.
  • the disparity vector is set to the initial (0, 0), and is performed in the order of the depth map encoding after encoding each view point as described in step (1);
  • step (3.3.2) the neighboring disparity vector of the reference frame in the time direction is obtained by detecting whether the disparity vector flag variable is 0, and if it is 0, detecting the CTR position of the corresponding position of the reference frame, and if the disparity vector can be detected,
  • the disparity vector flag variable is set to 1;
  • Step (3.3.3) performing parallax compensation prediction of disparity vector of spatial position, marking the group of A1 and A0 as group 1, and the group of B1 and B0 as group 2.
  • the spatial direction direction intra-frame adjacent block disparity compensation prediction disparity vector acquisition method is: detecting whether the disparity vector flag variable is 0, and if it is 0, judging A1 in group 1, if the disparity compensation prediction disparity vector is found in A1, and then continuing to search for A0, judging Whether the disparity compensated prediction disparity vector is found, and if the disparity compensated prediction disparity vector is found, the disparity compensated prediction disparity vector in A1 and the disparity compensated predicted disparity vector in A0 are combined based on the packet combining ratio table (Table 3) in step (3.2).
  • the A1 position disparity vector is used and the flag variable is set to 1.
  • the parallax compensated prediction disparity vector in A1 is (5, 4)
  • the A0 disparity compensated prediction disparity vector is (4, 4)
  • Multiply the A1 adoption rate by 68.6%, (4, 4) multiply the A0 adoption rate by 31.4%, and add the new A1 position disparity vector (5, 4) and use (where 5 ⁇ 68.6% + 4 ⁇ 31.4% rounding equals 5,4 ⁇ 68.6%+4 ⁇ 31.4% is rounded up to 4).
  • A0 does not find the disparity compensated prediction disparity vector
  • the parallax compensated prediction disparity vector in A1 is used, and the search program terminates the bounce, and the subsequent position is no longer traversed. If the disparity compensated prediction disparity vector is not found in A1, skip the A0 and directly detect the disparity compensated prediction disparity vector of the B1 position. If the disparity compensated prediction disparity vector is found in B1, determine whether the B0 position finds the disparity compensated prediction disparity vector, if the disparity is obtained.
  • Compensating the predicted disparity vector finds the disparity compensated disparity vector in B1 and the disparity compensated disparity vector in B0 are combined into a disparity vector of the B2 position based on the packet combining ratio table (Table 3) in step (3.2), and the disparity vector is adopted.
  • the flag variable is set to 1, for example, the disparity compensated prediction disparity vector in B1 is (8, 9), and the B0 disparity compensated predicted disparity vector is (7, 8), then multiplied by B1 adoption rate by 23.1%.
  • Step (3.3.4) performing motion compensated prediction disparity vector detection of the spatial position, marking the group of A1 and A0 as group 1, and the group of B1 and B0 as group 2.
  • the spatial direction direction intra-frame adjacent block motion compensation prediction disparity vector acquisition method is: detecting whether the disparity vector flag variable is 0, and if it is 0, judging A1 in group 1, if the motion compensation prediction disparity vector is found in A1, and then continuing to search A0, judging Whether the motion compensated prediction disparity vector is found, and if the motion compensated prediction disparity vector is found, the motion compensated prediction disparity vector in A1 and the motion compensated predicted disparity vector in A0 are based on the step (3.2).
  • the group combination ratio table (Table 3) is combined and used as the A1 position disparity vector, and the flag variable is set to 1, for example, the motion compensation prediction disparity vector in A1 is (5, 4), and the A0 motion compensation prediction disparity vector is (4, 4), multiply (5, 4) by A1 adoption rate of 68.6%, (4, 4) multiplied by A0 adoption rate of 31.4% and add up to obtain new A1 position disparity vector (5, 4) and adopt (of which 5 ⁇ 68.6%+4 ⁇ 31.4% rounding equals 5,4 ⁇ 68.6%+4 ⁇ 31.4% rounding equals 4). If A0 does not find the motion compensation prediction disparity vector, the motion compensation prediction disparity vector in A1 is used to find the program termination.
  • the motion compensated prediction disparity vector finds the motion compensated prediction disparity vector in B1 and the motion compensated predicted disparity vector in B0 in combination with the disparity vector of the B2 position based on the packet combining scale table (Table 3) in step (3.2), and the parallax is employed.
  • the vector flag variable is set to 1, for example, the motion compensated prediction disparity vector in B1 is (8, 9), and the B0 motion compensated prediction disparity vector is (7, 8), then multiplied by B1 using rate (2, 9). %, (7,8) multiplied by B0 adoption rate of 76.9% and then added to obtain a new B1 position disparity vector (7,8) and adopted (where 8 ⁇ 23.1%+7 ⁇ 76.9% is rounded up to 7,9 ⁇ 23.1%) +8 ⁇ 76.9% rounding equals 8). If the motion compensation prediction disparity vector is not found in B1 of group 2, skip step (3.3.4);
  • Step (3.3.5) judge the depth optimization flag, and if the flag bit is 1, use the depth optimization operation.
  • the depth optimization operation (defined by DoNBDV in Reference 1) is prior art.
  • the texture block to which the final disparity vector is located for example, the final disparity vector is (3, 4), and the position of the current texture block is (1, 2), then the position of the texture block pointed to it is (4, 6) , (3+1 is 4, 4+2 is 6).
  • the depth block corresponding to the texture block pointed to by the final disparity vector is searched for the depth value of the pixel position of the 4 corners of the depth block.
  • this maximum value is converted into a depth-optimized disparity vector.
  • the conversion formula is:
  • f is the camera focal length value
  • l is the baseline distance
  • d max is the maximum depth value of the pixel position of the 4 corners of the search depth block.
  • the invention proposes a method, which changes the first searched disparity vector, which is regarded as the criterion of the final disparity vector, and changes the original method to reduce the number of search times by deleting the search candidate position to bring the coding quality. Insufficient, Under the premise of maintaining the efficiency of the existing fast algorithm (the decoding time is shortened to 97.1%, the encoding and virtual view synthesis time are unchanged), at least the coding quality is improved by 0.05%, and the test sequence is compared with the existing fast algorithm. 4 is shown.
  • 1 is a predictive coding structure diagram of 3D-HEVC
  • 2 is a coding sequence diagram of a texture map priority in 3D-HEVC
  • 3 is a candidate spatial coding block position map based on a disparity vector acquisition method of a neighboring block
  • 5 is a candidate spatial coding block position map based on a subtraction method
  • Figure 7 is a block diagram of a candidate spatial coding block of the method of the present invention.
  • Figure 8 is a block diagram of candidate time coded block locations of the method of the present invention.
  • Figure 9 is a flow chart of the method of the present invention.
  • the technical solution adopted by the present invention is:
  • a method for acquiring adjacent disparity vectors in auxiliary view coding in multi-texture multi-depth video coding firstly extracting candidate bit spaces and time coding unit positions adjacent to a current coding unit from coded coding units in a current coded view point Information of the disparity vector adoption rate; then, by deleting the candidate space adjacent to the current coding unit and the least searched position in the temporal coding unit position, grouping the adjacent candidate space and the time coding unit position, and searching for the parallax thereof
  • the vectors are combined into a final disparity vector according to the ratio of adoption rates.
  • Changing the first searched disparity vector is regarded as the criterion of the final disparity vector, which changes the original method to reduce the number of search times by reducing the search candidate position, which can bring the lack of coding quality. At least the coding quality is improved with the efficiency of the fast algorithm.
  • n frames (n is greater than or equal) for the 3D-HEVC international standard test video sequence Newspaper_CC, GT fly, Undo dancer, Poznan_Hall2, Poznan_Street, Kendo, Balloons, respectively.
  • the main view and the auxiliary view of 40 and less than or equal to 60 are encoded, and each view is encoded by using texture map coding and then performing depth map coding;
  • the all-position search method is adopted for the candidate spatial and temporal coding unit positions adjacent to the current coding unit, and the candidate vector space of the current coding unit and the disparity vector adoption rate of the temporal coding unit position are simultaneously extracted.
  • the information includes: detection of adjacent candidate temporal coding unit positions CTR and BR and whether adjacent candidate spatial coding unit positions A1, B1, B0, A0, B2 contain disparity vectors or motion compensated prediction disparity vectors, and Performing a difference sum of squares (SSD, Sum of Squared Difference) on the disparity vector found by all adjacent candidate spatial and temporal coding unit positions of the current coding unit or the disparity reference frame pointed by the motion compensated prediction disparity vector information, and the difference is squared
  • SSD Sum of Squared Difference
  • the least-searched position in the candidate space adjacent to the current coding unit and the position of the time coding unit is deleted
  • the adjacent candidate space and time coding unit locations are grouped, including the following steps:
  • the adjacent candidate space and time coding unit positions remaining after the deletion are grouped, and the reference search position includes the A1, B1, B0, and A0 of the spatial position and the CRT of the time reference position by using the step (2.1).
  • the grouping operation is for coding units of adjacent positions, where B1 and B0 are adjacent, and A1 and A0 are adjacent, then B1 and adjacent B0 are grouped, and A1 and adjacent A0 are divided into another group.
  • the combination ratio of A1 and A0 and B1 and B0 after grouping is shown in Table 3;
  • the current coding unit performs adjacent disparity vectors on the adjacent candidate spatial and temporal coding unit search positions and group information.
  • the search and the final disparity vector synthesis operation including the following steps:
  • the disparity vector is set to an initial (0, 0), and is performed in the order of depth map encoding after encoding each view point as described in step (1.1);
  • the neighboring disparity vector of the reference frame in the time direction is obtained by detecting whether the disparity vector flag variable is 0, and if it is 0, detecting the CTR position of the corresponding position of the reference frame. If the disparity vector can be detected, the disparity vector flag variable is set. Is 1;
  • the disparity vector detection is performed by marking the group of A1 and A0 as group 1, and the group of B1 and B0 as group 2.
  • the spatial direction direction intra-frame adjacent block disparity compensation prediction disparity vector acquisition method is: detecting whether the disparity vector flag variable is 0, and if it is 0, judging A1 in group 1, if the disparity compensation prediction disparity vector is found in A1, and then continuing to search for A0, judging Whether the disparity compensated prediction disparity vector is found, and if the disparity compensated prediction disparity vector is found, the disparity compensated prediction disparity vector in A1 and the disparity compensated predicted disparity vector in A0 are combined based on the packet combining ratio table (Table 3) in step (2.2).
  • the A1 position disparity vector is used, and the flag variable is set to 1. If A0 does not find the disparity compensated prediction disparity vector, the parallax compensated prediction disparity vector in A1 is used, and the search program terminates the bounce, and the subsequent position is no longer traversed. If the disparity compensated prediction disparity vector is not found in A1, skip the A0 and directly detect the disparity compensated prediction disparity vector of the B1 position. If the disparity compensated prediction disparity vector is found in B1, determine whether the B0 position finds the disparity compensated prediction disparity vector, if the disparity is obtained.
  • Compensating the predicted disparity vector finds the parallax-compensated predicted disparity vector in B1 and the disparity-compensated predicted disparity vector in B0 by combining the disparity vector of the B2 position based on the packet combining ratio table (Table 3) in step (2.2), and adopting the disparity vector
  • the flag variable is set to 1. If the disparity compensated prediction disparity vector is not found in B1 of group 2, skip step (3.3);
  • Spatial direction direction intra-frame adjacent block motion compensation prediction disparity vector acquisition mode is detection Whether the disparity vector flag variable is 0, if it is 0, it judges A1 in group 1.
  • the motion compensation prediction disparity vector is found in A1, and then continues to search A0, it is judged whether the motion compensation prediction disparity vector is found, and if the motion compensation prediction disparity vector is found,
  • the motion compensated prediction disparity vector in A1 and the motion compensated prediction disparity vector in A0 are combined and used as the A1 position disparity vector based on the packet combining scale table (Table 3) in step (2.2), and the flag variable is set to 1, if A0 is not found.
  • the motion compensated prediction disparity vector uses the motion compensated prediction disparity vector in A1 to find that the program terminates the bounce and the subsequent position is no longer traversed.
  • the motion compensated prediction disparity vector If the motion compensated prediction disparity vector is not found in A1, skip the A0 and directly detect the motion compensated prediction disparity vector of the B1 position. If the motion compensated prediction disparity vector is found in B1, determine whether the B0 position finds the motion compensated prediction disparity vector.
  • the motion compensated prediction disparity vector finds the motion compensated prediction disparity vector in B1 and the motion compensated predicted disparity vector in B0 in combination with the disparity vector of the B2 position based on the packet combining scale table (Table 3) in step (2.2), and the parallax is employed.
  • the vector flag variable is set to 1. If the motion compensation prediction disparity vector is not found in B1 of group 2, skip step (3.4);
  • f is the camera focal length value
  • l is the baseline distance
  • d max is the maximum depth value of the pixel position of the 4 corners of the search depth block.
  • the multi-texture multi-depth 3D-HEVC video sequence is first read.
  • the input video sequence needs to be in YUV format.
  • After the computer reads the video data first use HTM8.0 multi-texture multi-depth video 3D-HEVC encoding software. Encoding both the primary view and the secondary view of the sequence, and extracting the information of the disparity vector adoption rate of the candidate space adjacent to the current coding unit and the position of the temporal coding unit while encoding, and positioning the adjacent candidate space and the time coding unit The grouping is performed, and then the method mentioned in the present invention is called to complete the specific encoding work for the multi-texture multi-depth video.
  • FIG. 6 is a flow chart of the method of the present invention, and the specific implementation steps are as follows:
  • the video sequence used in the experiment is a multi-texture multi-depth international standard test video sequence Newspaper_CC, GT fly, Undo dancer, Poznan_Hall2, Poznan_Street, Kendo, Balloons sequence, respectively selecting the first n frames (n is greater than or equal to 40 and less than or equal to 60).
  • the natural number, for example, n is 40) the primary and secondary viewpoints are encoded.
  • the HTM 8.0 software is used to perform an all-position search method on the candidate spatial and temporal coding unit positions adjacent to the current coding unit to obtain the information of the disparity vector adoption rate of the candidate space and the temporal coding unit position adjacent to the current coding unit, and then utilize
  • the disparity vector adoption rate information deletes the candidate space adjacent to the current coding unit and the least searched position in the temporal coding unit position, groups the adjacent candidate space and the time coding unit position, and uses the searched disparity vector according to the adoption rate.
  • the ratios are combined into a final disparity vector, and finally the method of the present invention is used to acquire the auxiliary view neighboring disparity vectors of the remaining sequence, and the first searched disparity vector is changed as the criterion of the final disparity vector, and the change is changed.
  • the original method reduces the coding quality by reducing the search candidate position and reducing the number of search times, which can improve the coding quality while maintaining the efficiency of the original fast algorithm.
  • the first step using the video encoding software HTM8.0 to select the first n frames (n is the 3D-HEVC international standard test video sequence Newspaper_CC, GT fly, Undo dancer, Poznan_Hall2, Poznan_Street, Kendo, Balloons respectively)
  • the primary view and the auxiliary view are encoded by a natural number greater than or equal to 40 and less than or equal to 60, and the depth map coding is performed after encoding each view by using texture map coding;
  • the all-position search method is adopted for the candidate spatial and temporal coding unit positions adjacent to the current coding unit, and the candidate vector space of the current coding unit and the disparity vector adoption rate of the temporal coding unit position are simultaneously extracted.
  • the information includes: detection of adjacent candidate temporal coding unit positions CTR and BR and whether adjacent candidate spatial coding unit positions A1, B1, B0, A0, B2 contain disparity vectors or motion compensated prediction disparity vectors, and Performing a difference sum of squares (SSD, Sum of Squared Difference) on the disparity vector found by all adjacent candidate spatial and temporal coding unit positions of the current coding unit or the disparity reference frame pointed by the motion compensated prediction disparity vector information, and the difference is squared
  • SSD Sum of Squared Difference
  • Step 2 Using the parallax of the candidate space adjacent to the current coding unit and the position of the time coding unit obtained in the previous step
  • the information of the vector adoption rate provides a basis for group search: firstly, the deletion operation is performed for the adjacent candidate time and spatial coding unit positions, and the coding unit position with the smallest disparity vector adoption rate is deleted.
  • the usage rate statistics of each location are shown in Table 1 and Table 2;
  • the third step grouping the adjacent candidate space and time coding unit positions remaining after the previous step is deleted, and obtaining the C1 of the reference search position including the spatial positions A1, B1, B0, and A0 and the time reference position by the second step. , as shown in Figure 7 and Figure 8.
  • the grouping operation is for coding units of adjacent positions, where B1 and B0 are adjacent, and A1 and A0 are adjacent, then B1 and adjacent B0 are grouped, and A1 and adjacent A0 are divided into another group.
  • the combination ratio of A1 and A0 and B1 and B0 after grouping is shown in Table 3;
  • Step 4 Set the acquisition disparity vector flag variable.
  • the disparity vector flag variable characterizes whether the disparity vector is obtained. If it has been acquired, it is 1 and if it is not, it is 0.
  • the disparity vector is set to the initial (0, 0), and is performed in the order of depth map encoding after encoding each view point as described in the first step;
  • Step 5 The neighboring disparity vector of the reference frame in the time direction is obtained by detecting whether the disparity vector flag variable is 0, and if it is 0, detecting the CTR position of the corresponding position of the reference frame, and if the disparity vector can be detected, the disparity vector flag The variable is set to 1;
  • Step 6 Perform parallax compensation prediction of disparity vector detection of spatial position, mark the group of A1 and A0 as group 1, and the group of B1 and B0 as group 2.
  • the spatial direction direction intra-frame adjacent block disparity compensation prediction disparity vector acquisition method is: detecting whether the disparity vector flag variable is 0, and if it is 0, judging A1 in group 1, if the disparity compensation prediction disparity vector is found in A1, and then continuing to search for A0, judging Whether the disparity compensated prediction disparity vector is found, and if the disparity compensated prediction disparity vector is found, the disparity compensated prediction disparity vector in A1 and the disparity compensated disparity vector base in A0 are combined into a grouping ratio table (Table 3) in the third step as A1.
  • Table 3 grouping ratio table
  • the position disparity vector is used, and the flag variable is set to 1. If A0 does not find the disparity compensated prediction disparity vector, the parallax compensated prediction disparity vector in A1 is used, and the search program terminates the bounce, and the subsequent position is no longer traversed. If the disparity compensated prediction disparity vector is not found in A1, skip the A0 and directly detect the disparity compensated prediction disparity vector of the B1 position. If the disparity compensated prediction disparity vector is found in B1, determine whether the B0 position finds the disparity compensated prediction disparity vector, if the disparity is obtained.
  • Compensating the predicted disparity vector finds the disparity compensated disparity vector in B1 and the disparity compensated disparity vector in B0 are combined into a disparity vector of the B2 position based on the packet combining scale table (Table 3) in the third step, and the disparity vector flag is adopted.
  • the variable is set to 1. Skip this step if B1 in group 2 does not find the disparity compensated prediction disparity vector;
  • Step 7 Perform motion compensation prediction of parallax vector detection of spatial position, and mark the group of A1 and A0 as group 1,
  • the groups of B1 and B0 are labeled as group 2.
  • the spatial direction direction intra-frame adjacent block motion compensation prediction disparity vector acquisition method is: detecting whether the disparity vector flag variable is 0, and if it is 0, judging A1 in group 1, if the motion compensation prediction disparity vector is found in A1, and then continuing to search A0, judging Whether the motion compensated prediction disparity vector is found, and if the motion compensated prediction disparity vector is found, the motion compensated prediction disparity vector in A1 and the motion compensated predicted disparity vector in A0 are combined into A1 based on the packet combining ratio table (Table 3) in the third step.
  • the position disparity vector is used, and the flag variable is set to 1. If A0 does not find the motion compensation prediction disparity vector, the motion compensation prediction disparity vector in A1 is used, and the search program terminates the bounce, and the subsequent position is no longer traversed. If the motion compensated prediction disparity vector is not found in A1, skip the A0 and directly detect the motion compensated prediction disparity vector of the B1 position. If the motion compensated prediction disparity vector is found in B1, determine whether the B0 position finds the motion compensated prediction disparity vector.
  • the motion compensated prediction disparity vector finds the motion compensated prediction disparity vector in B1 and the motion compensated predicted disparity vector in B0 in combination with the disparity vector of the B2 position based on the packet combining ratio table (Table 3) in the third step, and the disparity vector is adopted.
  • the flag variable is set to 1. If the motion compensation prediction disparity vector is not found in B1 of group 2, skip this step;
  • Step 8 Determine the depth optimization flag. If the flag is 1, use the depth optimization operation.
  • the depth optimization operation (defined by DoNBDV in Reference 1) is prior art. First look at the texture block that the final disparity vector is looking for. The depth block corresponding to the texture block pointed to by the final disparity vector is searched for the depth value of the pixel position of the 4 corners of the depth block. Finally, this maximum value is converted into a depth-optimized disparity vector.
  • the conversion formula is:
  • f is the camera focal length value
  • l is the baseline distance
  • d max is the maximum depth value of the pixel position of the 4 corners of the search depth block.
  • the invention changes the first searched disparity vector, which is regarded as the criterion of the final disparity vector, and changes the deficiencies of the original method to reduce the search quality by deleting the search candidate position and can maintain the coding quality.
  • the efficiency of the original fast algorithm is at least improved under the premise of the efficiency of the original fast algorithm.
  • the experimental results of the existing fast algorithm are shown in Table 4.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)
  • Testing, Inspecting, Measuring Of Stereoscopic Televisions And Televisions (AREA)

Abstract

一种多纹理多深度视频编码中辅助视点编码中相邻视差矢量获取方法，属于3D-HEVC视频编码技术领域，其特征在于，改变第1个被搜索到的视差矢量即被当作最终视差矢量的准则，通过删除当前编码单元相邻的候选空间和时间编码单元位置中最少被搜索的位置，将相邻的候选空间和时间编码单元位置进行分组，将其搜索到的视差矢量按照采用率的比例组合为最终视差矢量的方法，能在保持了原有快速算法的效率前提下提升编码质量。改变了原有方法通过删减搜索候选位置减少搜索次数的方式给编码质量带来的不足，能在保持了现有快速算法的效率前提下(解码时间缩短为原有97.1％，编码和虚拟视点合成时间不变)至少提升编码质量0.05％。

Description

一种面向多纹理多深度视频的相邻视差矢量获取方法 技术领域
本发明涉及基于3D-HEVC的视频编码技术，具体涉及一种多纹理多深度视频编码中辅助视点编码中相邻视差矢量获取方法。
背景技术
近二十年来，视频广播技术发生了重大变革，从上世纪的模拟电视到数字电视、高清数字电视，乃至现在的3D电视，视频技术随着人们生活水平的提高不断发展进步。当今世界，人们已经不再满足于传统的二维视频带来的视觉感受，具有临场感和交互性的三维立体视频逐渐成为多媒体信息产业的热门话题。MPEG的3DG工作组开始探索基于高效视频编码(HEVC，High Efficiency Video Coding)的多深度多纹理三维视频格式数据的压缩编码方法，2012年7月，VCEG和MPEG又共同成立JCT-3V小组，制定HEVC的3D视频编码扩展标准。并于2013年提出了构建基于HEVC的3D编码标准3D-HEVC。在3D-HEVC标准中采用的多深度多纹理视频是目前最有效的三维视频表现方式：多深度多纹理视频是由多台(通常拍摄3纹理3深度)相邻摄像机从不同角度对同一场景进行拍摄得到的多路具有细微视角差异的视频集合；加入多深度信息的多纹理视频可以更加细致、全面地描述三维场景信息，有利于三维视频终端在较大视角范围内，采用基于深度的虚拟视点绘制技术生成任意视角的高质量虚拟视图，并同时提供双目视差感与运动视差感，从而提供给用户身临其境的观看体验。
由于多深度多纹理视频数据量庞大，必须对其进行高效压缩编码。所以在3D-HEVC引入的视频编码标准中采用8×8到64×64的四叉树预测单元结构、4×4到32×32的变换单元尺寸、多角度的36种帧内预测模式、自适应环路滤波等多项新技术。同时针对3D-HEVC的多深度多纹理视频编码结构，引入了视点间的多参考帧，帧间预测编码模式的概念延伸为同一视点时间方向的运动预测编码模式和相邻视点方向的视差预测编码模式，使得其计算复杂度进一步更高。
以包含3纹理3深度的多纹理多深度视频为例，如图1所示，水平方向为时间方向，垂直方向为视点方向。在时间方向采用分层B帧结构来消除时间方向的冗余，在视点间方 向采用I-P-P结构消除试点间冗余信息。主视点只能利用本视点内的编码帧作为参考帧，辅助视点除了利用本视点内已编码的帧作为参考帧之外还可以利用主视点的编码帧作为参考帧。对于每一个视点，包含相对应的8比特表征的深度图。在3D-HEVC中，首先对主视点纹理图进行编码，然后对主视点深度图进行编码，然后依次对辅助视点进行纹理图和深度图编码。因为深度图需要纹理图的编码信息进行编码，同时辅助视点编码需要利用主视点的编码信息进行编码，叫做纹理图编码优先编码顺序，如图2所示。
视差矢量的获取是多纹理多深度的3D-HEVC视频编码技术中的一项关键技术，在视点间运动补偿预测和视点间残差预测中都有广泛的使用。视差矢量代表同一时刻内不同的视频帧之间的差异，在现有3D-HEVC标准中，主视点中的视差矢量可对辅助视点的预测单元进行运动补偿。在纹理图编码优先编码顺序中，当对辅助视点中的预测单元进行编码时，其使用的视差矢量不能从对应的深度图中计算得到，因为其对应深度图还未输入编码单元。
传统的视差矢量获取方法是通过块估计和块匹配的方法得到，在解码端也需要相关的信息来进行解码操作。如果这些信息在码流中传输，额外的传输比特将产生。为避免这一情况的产生，现有的3D-HEVC中引入了从已编码的纹理图信息估计深度图的技术。为了得到深度图，主视点和辅助视点之间的视差信息往往被转化为深度图信息，计算出的深度图信息可以用于转化为主视点和其它辅助视点深度图信息。在此过程中所估计的深度图中最大深度值将被转换为视差矢量,此过程被称为基于深度图的视差矢量转换。
由于基于深度图的视差矢量转换过程计算量庞大，为减少计算复杂度3D-HEVC又引入了简化的视差矢量获取算法，称为基于相邻块的视差矢量获取方法。基于相邻块的视差矢量获取方法按照预设顺序对候选空间和时间编码块位置进行搜索，判断是否含有视差矢量信息来获取当前块的视差矢量，空间和时间编码块位置如图3和图4所示。如果搜索到的预测单元中使用了视差补偿预测或者视差运动矢量补偿预测技术，则表示预测单元中含有视差信息，第1个被搜索到的视差矢量被用于视点间运动补偿预测和视点间残差预测的过程。搜索预设顺序为：首先搜索时间参考帧中位置CRT和BR，然后搜索空间参考帧位置A1，B1，B0，A0和B2，最后再搜索上述空间参考帧位置中的运动矢量补偿情况。基于相邻块的视差矢量获取方法比基于深度图的视差矢量转换方法至少节省8％以上的时间。
基于相邻块的视差矢量获取方法出现后，3D-HEVC又利用基于深度图的相邻块视差矢量获取方法来改进所获取的相邻视差矢量。基于深度图的相邻块视差矢量获取方法将已编码的主视点深度图用于修正初始获取的视差矢量。在获取原始的相邻块的视差矢量后，再利用相关主视点的深度图的最大深度值来修正进而得到最终的视差矢量。
为进一步节省传输比特加速获取相邻视差矢量的过程，一些研究机构进行了基于相邻块视差矢量获取的快速算法的研究。华为海思公司提出跳过预测单元中重叠位置的搜索方法，高通公司提出以编码单元为最小算法执行单位，三星公司提出删减空间和时间参考帧中搜索位置的方法(如图5和图6所示)。此外，一些研究机构提出研究可变编码工具来改变3D-HEVC的编码顺序。
综上所述，现有的基于相邻块的视差矢量获取的改进算法都将研究重心放到了删减搜索候选位置减少搜索次数。其主要问题在于，第1个被搜索到的视差矢量即被当作最终的视差矢量，搜索停止，剩余搜索位置中仍然有可能含有可被利用甚至更好的视差矢量在还没被搜索之前就被整个获取过程中止掉了。所以本发明基于3D-HEVC标准，提出一种方法，改变第1个被搜索到的视差矢量即被当作最终视差矢量的准则，通过删除当前编码单元相邻的候选空间和时间编码单元位置中最少被搜索的位置，同时将相邻的候选空间和时间编码单元位置进行分组，将其搜索到的视差矢量按照采用率的比例组合为最终视差矢量的方法，能在保持了原有快速算法的效率前提下提升编码质量。
发明内容
本发明的目的在于，基于3D-HEVC标准，提出一种方法，改变现有基于相邻块的视差矢量获取方法中第1个被搜索到的视差矢量即被当作最终视差矢量的准则，通过删除当前编码单元相邻的候选空间和时间编码单元位置中最少被搜索的位置，同时将相邻的候选空间和时间编码单元位置进行分组，将其搜索到的视差矢量按照采用率的比例组合为最终视差矢量的方法，能在保持了原有快速算法的效率前提下提升编码质量。
本发明的特征在于，是在计算机中依次按以下步骤仿真实现的：
步骤(1)，计算机初始化：
设置：按照通用测试条件对所选定序列的主视点以及当前辅助视点进行编码，编码软 件采用3D-HEVC视频编码软件HTM8.0版本作为编码平台。按照通用测试条件规定的纹理图和对应深度图量化参数分为第1组(40，45)，第2组(35，42)，第3组(30，39)，第4组(25，34)，其中括号中前面的数字代表纹理图量化参数，后面为深度图量化参数，共4组。
以YUV为格式的3D-HEVC国际标准测试视频序列包括：Newspaper_CC(300帧)、GT fly(250帧)、Undo dancer(250帧)，Poznan_Hall2(200帧)，Poznan_Street(250帧)，Kendo(300帧)，Balloons(300帧)，每个测试序列包含3个纹理序列，以及对应3个深度序列。各个测试序列进行编码时输入编码器的3个视点顺序分别为：4-2-6，5-9-1，5-1-9，6-7-5，4-5-3，3-1-5，3-1-5，在对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行。例如，对Newspaper_CC序列进行编码时，按照通用测试条件纹理图和对应深度图量化参数组(40，45)初始化后，在计算机上运行的HTM编码软件首先读取第1帧编号为4的主视点序列纹理图进行编码，再读取第1帧编号为4的主视点深度图进行编码，主视点第1帧编码完成后，再读取第1帧编号为2的辅助视点序列纹理图进行编码，完成后读取第1帧编号为2的辅助视点序列深度图进行编码，最后读取第1帧编号为6的辅助视点序列纹理图进行编码，完成后读取第1帧编号为6的辅助视点深度图进行编码。由此完成3个视点序列的第1帧编码，依次读取到300帧完成Newspaper_CC序列在量化参数组(40，45)下的编码。然后按照通用测试条件纹理图和对应深度图量化参数组(35，42)初始化后，在计算机上运行的HTM编码软件首先读取第1帧编号为4的主视点序列纹理图进行编码，再读取第1帧编号为4的主视点深度图进行编码，主视点第1帧编码完成后，再读取第1帧编号为2的辅助视点序列纹理图进行编码，完成后读取第1帧编号为2的辅助视点序列深度图进行编码，最后读取第1帧编号为6的辅助视点序列纹理图进行编码，完成后读取第1帧编号为6的辅助视点深度图进行编码。由此完成3个视点序列的第1帧编码，依次读取到300帧完成Newspaper_CC序列在量化参数组(35，42)下的编码。然后依次完成在Newspaper_CC序列在量化参数组(30，39)和(25，34)下的编码。又例如在对GT fly序列进行编码时，按照通用测试条件纹理图和对应深度图量化参数组(40，45)初始化后，在计算机上运行的HTM编码软件首先读取第1帧编号为5的主视点序列纹理图进行编码，再读取第1帧编号为5的主视点深度图进行编码，主视点第1帧编码完成后，再读取第1帧编 号为9的辅助视点序列纹理图进行编码，完成后读取第1帧编号为9的辅助视点序列深度图进行编码，最后读取第1帧编号为1的辅助视点序列纹理图进行编码，完成后读取第1帧编号为1的辅助视点深度图进行编码。由此完成3个视点序列的第1帧编码，依次读取到250帧完成GT fly序列在量化参数组(40，45)下的编码。然后按照通用测试条件纹理图和对应深度图量化参数组(35，42)初始化后，在计算机上运行的HTM编码软件首先读取第1帧编号为5的主视点序列纹理图进行编码，再读取第1帧编号为5的主视点深度图进行编码，主视点第1帧编码完成后，再读取第1帧编号为9的辅助视点序列纹理图进行编码，完成后读取第1帧编号为9的辅助视点序列深度图进行编码，最后读取第1帧编号为1的辅助视点序列纹理图进行编码，完成后读取第1帧编号为1的辅助视点深度图进行编码。由此完成3个视点序列的第1帧编码，依次读取到250帧完成GT fly序列在量化参数组(35，42)下的编码。然后依次完成在GT fly序列在量化参数组(30，39)和(25，34)下的编码；
步骤(2)，利用所述的视频编码软件HTM8.0对所述的3D-HEVC国际标准测试视频序列Newspaper_CC、GT fly、Undo dancer，Poznan_Hall2，Poznan_Street，Kendo，Balloons，分别选取前n帧(n为大于等于40并小于等于60的自然数)的所述主视点和辅助视点进行编码，对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行。在编码前n帧时，采用对当前编码单元相邻的候选空间和时间编码单元位置进行全位置搜索方法，并同时提取当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息，包括：对相邻的候选时间编码单元位置CTR和BR和对相邻的候选空间编码单元位置A1，B1，B0，A0，B2是否含有视差矢量或者运动补偿预测视差矢量的检测，以及通过对当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量或者运动补偿预测视差矢量信息指向的视差参考帧进行差值平方和(SSD，Sum of Squared Difference)计算，以差值平方和信息进行当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量采用率统计。例如，对Newspaper_CC序列在量化参数组(40，45)下前n帧(n取40)进行编码时，在计算机上运行的HTM编码软件，在编码其辅助视点时首先读取第1帧的辅助视点纹理图，再读取当前纹理图第1个编码块，搜索第1个编码块时间方向参考帧中对应的编码单元位置CTR和BR中检测是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有视差矢量，再搜索空间方向参考帧中编码单元 位置A1，B1，B0，A0，B2中是否含有运动补偿预测视差矢量，例如在第1个编码块中找到A0和A1都含有视差矢量，则由A0位置和A1位置视差矢量分别找到其在视差参考帧中的对应编码单元分别进行差值平方和计算，若A1位置差值平方和为最小，则记为A1位置采用1次。依次读取第1帧所有编码块完成第1帧辅助视点纹理图采用次数统计，并按照纹理图编码完成后再进行深度图编码的顺序读取前40帧完成序列Newspaper_CC在量化参数组(40，45)的位置的次数统计。然后对序列Newspaper_CC在量化参数组(35，42)下前n帧(n取40)进行编码，在计算机上运行的HTM编码软件，在编码其辅助视点时首先读取第1帧的辅助视点纹理图，再读取当前纹理图第1个编码块，搜索第1个编码块时间方向参考帧中对应的编码单元位置CTR和BR中检测是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有运动补偿预测视差矢量，例如在第1个编码块中找到A0和A1都含有视差矢量，则由A0位置和A1位置视差矢量分别找到其在视差参考帧中的对应编码单元分别进行差值平方和计算，若A1位置差值平方和为最小，则记为A1位置采用1次。依次读取第1帧所有编码块完成第1帧的辅助视点纹理图采用次数统计，并按照纹理图编码完成后再进行深度图编码的顺序读取前40帧完成序列Newspaper_CC在量化参数组(35，42)的位置的次数统计。然后依次完成在Newspaper_CC序列在量化参数组(30，39)和(25，34)下的次数统计，最后由各位置采用次数除以总次数得到Newspaper_CC序列各位置采用率统计。又例如，对GT fly序列在量化参数组(40，45)下前n帧(n取40)进行编码时，在计算机上运行的HTM编码软件，在编码其辅助视点时首先读取第1帧的辅助视点纹理图，再读取当前纹理图第1个编码块，搜索第1个编码块时间方向参考帧中对应的编码单元位置CTR和BR中检测是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有运动补偿预测视差矢量，例如在第1个编码块中找到A0和A1都含有视差矢量，则由A0位置和A1位置视差矢量分别找到其在视差参考帧中的对应编码单元分别进行差值平方和计算，若A1位置差值平方和为最小，则记为A1位置采用1次。依次读取第1帧所有编码块完成第1帧的辅助视点纹理图采用次数统计，并按照纹理图编码完成后再进行深度图编码的顺序读取前40帧完成序列 GT fly在量化参数组(40，45)的位置的次数统计。然后对序列GT fly在量化参数组(35，42)下前n帧(n取40)进行编码，在计算机上运行的HTM编码软件，在编码其辅助视点时首先读取第1帧的辅助视点纹理图，再读取当前纹理图第1个编码块，搜索第1个编码块时间方向参考帧中对应的编码单元位置CTR和BR中检测是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有视差矢量，再搜索空间方向参考帧中编码单元位置A1，B1，B0，A0，B2中是否含有运动补偿预测视差矢量，例如在第1个编码块中找到A0和A1都含有视差矢量，则由A0位置和A1位置视差矢量分别找到其在视差参考帧中的对应编码单元分别进行差值平方和计算，若A1位置差值平方和为最小，则记为A1位置采用1次。依次读取第1帧所有编码块完成第1帧的辅助视点纹理图采用次数统计，并按照纹理图编码完成后再进行深度图编码的顺序读取前40帧完成序列GT fly在量化参数组(35，42)的位置的次数统计。然后依次完成在GT fly序列在量化参数组(30，39)和(25，34)下的次数统计，最后由各位置采用次数除以总次数得到GT fly序列各位置采用率统计。其中各位置采用率统计结果如表1和表2所示；
Figure PCTCN2015077944-appb-000001
表1 候选空间单元位置采用率统计
Figure PCTCN2015077944-appb-000002
表2 候选时间编码单元位置采用率统计
步骤(3)，对所述多纹理多深度视频中除主视点以外的所有辅助视点的编码帧的编码单元按照对当前编码单元相邻的候选空间和时间编码单元位置分组搜索并依采用率的比例组合为最终视差矢量方法进行相邻视差矢量获取，其步骤如下：
步骤(3.1)，利用步骤(2)得到的当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息为分组搜索提供依据：首先针对相邻的候选时间和空间编码单元位置进行删除操作，删除视差矢量采用率最小的编码单元位置。例如，在Newspaper_CC序列中最后得到的各位置采用率统计结果为：空间方向参考帧中编码单元位置A1，B1，B0，A0，B2采用率为65.7％，0.9％，3.0％，30.1％，0.2％，时间方向参考帧中对应的编码单元位置CTR和BR采用率为76.6％，23.4％。最后空间方向参考帧中编码单元位置B2被删除，时间方向参考帧中对应的编码单元位置BR被删除。又例如，在GT fly序列中最后的各位置采用率统计为：空间方向参考帧中编码单元位置A1，B1，B0，A0，B2采用率为66.7％，2.1％，6.0％，24.2％，1.0％，时间方向参考帧中对应的编码单元位置CTR和BR采用率为78.5％，21.5％。最后空间方向参考帧中编码单元位置B2被删除，时间方向参考帧中对应的编码单元位置BR被删除。其余测试序列各位置根据采用率统计结果得出，空间方向参 考帧中编码单元位置B2被删除，时间方向参考帧中对应的编码单元位置BR被删除；
步骤(3.2),针对删除之后剩余的相邻的候选空间和时间编码单元位置进行分组，通过步骤(3.1)得到参考搜索位置包含空间位置的A1，B1，B0，和A0以及时间参考位置的CRT，如图7和图8所示。分组操作针对邻近位置的编码单元，其中B1和B0相邻，A1和A0相邻，则B1和相邻的B0分为一组，A1和相邻的A0分为另一组，例如：在Newspaper_CC序列中空间方向参考帧中编码单元位置B2被删除，则候选空间编码单元位置A1和B1编码单元位置和其相邻的A0和B0编码单元位置形成组。又例如：在GT fly序列中空间方向参考帧中编码单元位置B2被删除，则候选空间编码单元位置A1和B1编码单元位置和其相邻的A0和B0编码单元位置形成组。其余序列空间方向参考帧中编码单元位置B2被删除，则候选空间编码单元位置A1和B1编码单元位置和其相邻的A0和B0编码单元位置形成组。其中A1和A0以及B1和B0分组后的结合比例如表3所示；
Figure PCTCN2015077944-appb-000003
表3 候选空间编码单元位置分组结合率
步骤(3.3)，利用步骤(3.2)得到的当前编码单元对相邻的候选空间和时间编码单元搜索位置和分组信息进行相邻视差矢量的搜索和最终视差矢量的合成计算操作，其步骤如下：
步骤(3.3.1)，设置获取视差矢量标志变量，视差矢量标志变量表征视差矢量是否获取，已获取则为1，未获取则为0值。视差矢量设定为初始的(0，0)，并按照在步骤(1)所述对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
步骤(3.3.2)，在时间方向参考帧的相邻视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则检测参考帧对应位置的CTR位置，如果视差矢量可以检测得到，则视差矢量标志变量设为1；
步骤(3.3.3)，进行空间位置的视差补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2。空间方向帧内相邻块视差补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到视差补偿预测视差矢量，再继续搜索A0，判断是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将A1中的视差补偿预测视差矢量和A0中的视差补偿预测视差矢量基于步骤(3.2)中分组结合比例表(表3)结合为A1位置视差矢量并采用，标志变量设置为1，例如：A1中的视差补偿预测视差矢量为(5,4)，A0视差补偿预测视差矢量为(4,4)，则按照(5,4)乘以A1采用率68.6％，(4,4)乘以A0采用率31.4％再相加得到新的A1位置视差矢量(5,4)并采用(其中5×68.6％+4×31.4％四舍五入等于5，4×68.6％+4×31.4％四舍五入等于4)，若A0未找到视差补偿预测视差矢量，则采用A1中的视差补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历。若A1中未找到视差补偿预测视差矢量，则跳过A0直接检测B1位置的视差补偿预测视差矢量，若B1中找到视差补偿预测视差矢量，则判断B0位置是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将B1中视差补偿预测视差矢量和B0中的视差补偿预测视差矢量以基于步骤(3.2)中分组结合比例表(表3)结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1，例如：B1中的视差补偿预测视差矢量为(8,9)，B0视差补偿预测视差矢量为(7,8)，则按照(8,9)乘以B1采用率23.1％，(7,8)乘以B0采用率76.9％再相加得到新的B1位置视差矢量(7,8)并采用(其中8×23.1％+7×76.9％四舍五入等于7，9×23.1％+8×76.9％四舍五入等于8)。若组2中B1未找到视差补偿预测视差矢量则跳过步骤(3.3.3)；
步骤(3.3.4)，进行空间位置的运动补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2。空间方向帧内相邻块运动补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到运动补偿预测视差矢量，再继续搜索A0，判断是否找到运动补偿预测视差矢量，若运动补偿预测视差矢量找到则将A1中的运动补偿预测视差矢量和A0中的运动补偿预测视差矢量基于步骤(3.2)中分 组结合比例表(表3)结合为A1位置视差矢量并采用，标志变量设置为1，例如：A1中的运动补偿预测视差矢量为(5,4)，A0运动补偿预测视差矢量为(4,4)，则按照(5,4)乘以A1采用率68.6％，(4,4)乘以A0采用率31.4％再相加得到新的A1位置视差矢量(5,4)并采用(其中5×68.6％+4×31.4％四舍五入等于5，4×68.6％+4×31.4％四舍五入等于4)，若A0未找到运动补偿预测视差矢量，则采用A1中的运动补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历。若A1中未找到运动补偿预测视差矢量，则跳过A0直接检测B1位置的运动补偿预测视差矢量，若B1中找到运动补偿预测视差矢量，则判断B0位置是否找到运动补偿预测视差矢量，若视运动补偿预测视差矢量找到则将B1中运动补偿预测视差矢量和B0中的运动补偿预测视差矢量以基于步骤(3.2)中分组结合比例表(表3)结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1，例如：B1中的运动补偿预测视差矢量为(8,9)，B0运动补偿预测视差矢量为(7,8)，则按照(8,9)乘以B1采用率23.1％，(7,8)乘以B0采用率76.9％再相加得到新的B1位置视差矢量(7,8)并采用(其中8×23.1％+7×76.9％四舍五入等于7，9×23.1％+8×76.9％四舍五入等于8)。若组2中B1未找到运动补偿预测视差矢量则跳过步骤(3.3.4)；
步骤(3.3.5)，判断深度优化标志，若标志位为1采用深度优化操作。深度优化操作(参考文献1中DoNBDV定义)为已有技术。首先将面找的最终视差矢量其指向的纹理块，例如最终视差矢量为(3,4)，当前纹理块的位置为(1,2),则其指向的纹理块位置为(4,6),(3+1为4，4+2为6)。将最终视差矢量指向的纹理块对应的深度块，搜索深度块4个角的像素位置的深度值选取其最大值。最后将这个最大值转换为深度优化后的视差矢量，转换公式为：
Figure PCTCN2015077944-appb-000004
其中
Figure PCTCN2015077944-appb-000005
为深度优化后的视差矢量，f为摄像机焦距值，l为基线距离，dmax为搜索深度块4个角的像素位置的最大深度值，
Figure PCTCN2015077944-appb-000006
和
Figure PCTCN2015077944-appb-000007
为离摄像机最近和最远的位置坐标(其中f，l和
Figure PCTCN2015077944-appb-000008
都是摄像机参数表中的已有固定参数值)；
本发明提出一种方法，改变了第1个被搜索到的视差矢量即被当作最终视差矢量的准则，改变了原有方法通过删减搜索候选位置减少搜索次数的方式给编码质量带来的不足， 能在保持了现有快速算法的效率前提下(解码时间缩短为原有97.1％，编码和虚拟视点合成时间不变)至少提升编码质量0.05％，对测试序列对比现有快速算法实验结果如表4所示。
Figure PCTCN2015077944-appb-000009
表4 本方面方法的对测试序列对比HTM8.0版本中快速算法实验结果统计表
附图说明
图1是3D-HEVC的预测编码结构图；
图2是3D-HEVC中纹理图优先的编码顺序图；
图3是基于相邻块的视差矢量获取方法的候选空间编码块位置图；
图4是基于相邻块的视差矢量获取方法的候选时间编码块位置图；
图5是基于删减方法的候选空间编码块位置图；
图6是基于删减方法的候选时间编码块位置图；
图7是本发明方法的候选空间编码块位置图；
图8是本发明方法的候选时间编码块位置图；
图9是本发明方法的流程图；
具体实施方式
为解决上述技术问题，本发明采取的技术方案为：
一种多纹理多深度视频编码中辅助视点编码中相邻视差矢量获取方法，首先从当前编码视点中已编码的编码单元中提取当前编码单元相邻的候选空间和时间编码单元位置的 视差矢量采用率的信息；然后通过删除当前编码单元相邻的候选空间和时间编码单元位置中最少被搜索的位置，将相邻的候选空间和时间编码单元位置进行分组，将其搜索到的视差矢量按照采用率的比例组合为最终视差矢量。改变了第1个被搜索到的视差矢量即被当作最终视差矢量的准则，改变了原有方法通过删减搜索候选位置减少搜索次数的方式给编码质量带来的不足，能在保持了原有快速算法的效率前提下至少提升编码质量。
本发明中采用的提取当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息具体步骤如下：
1.1、利用所述的视频编码软件HTM8.0对所述的3D-HEVC国际标准测试视频序列Newspaper_CC、GT fly、Undo dancer，Poznan_Hall2，Poznan_Street，Kendo，Balloons，分别选取前n帧(n为大于等于40并小于等于60的自然数)的所述主视点和辅助视点进行编码，对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
在编码前n帧时，采用对当前编码单元相邻的候选空间和时间编码单元位置进行全位置搜索方法，并同时提取当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息，包括：对相邻的候选时间编码单元位置CTR和BR和对相邻的候选空间编码单元位置A1，B1，B0，A0，B2是否含有视差矢量或者运动补偿预测视差矢量的检测，以及通过对当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量或者运动补偿预测视差矢量信息指向的视差参考帧进行差值平方和(SSD，Sum of Squared Difference)计算，以差值平方和信息进行当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量采用率统计。最后各位置采用率统计结果如表1和表2所示；
在本发明所提供的多纹理多深度视频编码中辅助视点编码中相邻视差矢量获取方法中，所述通过删除当前编码单元相邻的候选空间和时间编码单元位置中最少被搜索的位置，将相邻的候选空间和时间编码单元位置进行分组，包括下述步骤：
2.1、利用得到的当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息为分组搜索提供依据：首先针对相邻的候选时间和空间编码单元位置进行删除操作，删除视差矢量采用率最小的编码单元位置。其中各位置采用率统计如表1和表2所示；
2.2、其次针对删除之后剩余的相邻的候选空间和时间编码单元位置进行分组，通过步骤(2.1)得到参考搜索位置包含空间位置的A1，B1，B0，和A0以及时间参考位置的CRT， 如图7和图8所示。分组操作针对邻近位置的编码单元，其中B1和B0相邻，A1和A0相邻，则B1和相邻的B0分为一组，A1和相邻的A0分为另一组。其中A1和A0以及B1和B0分组后的结合比例如表3所示；
本发明所提供的多纹理多深度视频编码中辅助视点编码中相邻视差矢量获取方法中，所述将当前编码单元对相邻的候选空间和时间编码单元搜索位置和分组信息进行相邻视差矢量的搜索和最终视差矢量的合成操作，包括下述步骤：
3.1、设置获取视差矢量标志变量，视差矢量标志变量表征视差矢量是否获取，已获取则为1，未获取则为0值。视差矢量设定为初始的(0，0)，并按照在步骤(1.1)所述对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
3.2、在时间方向参考帧的相邻视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则检测参考帧对应位置的CTR位置，如果视差矢量可以检测得到，则视差矢量标志变量设为1；
3.3、进行空间位置的视差补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2。空间方向帧内相邻块视差补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到视差补偿预测视差矢量，再继续搜索A0，判断是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将A1中的视差补偿预测视差矢量和A0中的视差补偿预测视差矢量基于步骤(2.2)中分组结合比例表(表3)结合为A1位置视差矢量并采用，标志变量设置为1，若A0未找到视差补偿预测视差矢量，则采用A1中的视差补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历。若A1中未找到视差补偿预测视差矢量，则跳过A0直接检测B1位置的视差补偿预测视差矢量，若B1中找到视差补偿预测视差矢量，则判断B0位置是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将B1中视差补偿预测视差矢量和B0中的视差补偿预测视差矢量以基于步骤(2.2)中分组结合比例表(表3)结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1。若组2中B1未找到视差补偿预测视差矢量则跳过步骤(3.3)；
3.4、进行空间位置的运动补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2。空间方向帧内相邻块运动补偿预测视差矢量获取方式为，检测 视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到运动补偿预测视差矢量，再继续搜索A0，判断是否找到运动补偿预测视差矢量，若运动补偿预测视差矢量找到则将A1中的运动补偿预测视差矢量和A0中的运动补偿预测视差矢量基于步骤(2.2)中分组结合比例表(表3)结合为A1位置视差矢量并采用，标志变量设置为1，若A0未找到运动补偿预测视差矢量，则采用A1中的运动补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历。若A1中未找到运动补偿预测视差矢量，则跳过A0直接检测B1位置的运动补偿预测视差矢量，若B1中找到运动补偿预测视差矢量，则判断B0位置是否找到运动补偿预测视差矢量，若视运动补偿预测视差矢量找到则将B1中运动补偿预测视差矢量和B0中的运动补偿预测视差矢量以基于步骤(2.2)中分组结合比例表(表3)结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1。若组2中B1未找到运动补偿预测视差矢量则跳过步骤(3.4)；
3.5、判断深度优化标志，若标志位为1采用深度优化操作。深度优化操作(参考文献1中DoNBDV定义)为已有技术。首先将面找的最终视差矢量其指向的纹理块。将最终视差矢量指向的纹理块对应的深度块，搜索深度块4个角的像素位置的深度值选取其最大值。最后将这个最大值转换为深度优化后的视差矢量，转换公式为：
Figure PCTCN2015077944-appb-000010
其中
Figure PCTCN2015077944-appb-000011
为深度优化后的视差矢量，f为摄像机焦距值，l为基线距离，dmax为搜索深度块4个角的像素位置的最大深度值，
Figure PCTCN2015077944-appb-000012
和
Figure PCTCN2015077944-appb-000013
为离摄像机最近和最远的位置坐标(其中f，l和
Figure PCTCN2015077944-appb-000014
都是摄像机参数表中的已有固定参数值)；
在实际的使用中，首先读入多纹理多深度3D-HEVC视频序列，输入的视频序列需为YUV格式，计算机读入视频数据后，首先利用HTM8.0多纹理多深度视频3D-HEVC编码软件对序列的主视点和辅助视点都进行编码，并在编码的同时提取当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息，将相邻的候选空间和时间编码单元位置进行分组，然后调用本发明中提到的方法来完成对多纹理多深度视频具体的编码工作。
本发明针对多纹理多深度视频辅助视点的所有帧的宏块，设计辅助视点编码中相邻视差矢量获取方法。图6是本发明方法流程图，其具体实施步骤如下：
实验中使用的视频序列为多纹理图多深度国际标准测试视频序列Newspaper_CC、GT fly、Undo dancer，Poznan_Hall2，Poznan_Street，Kendo，Balloons序列，分别选取前n帧(n为大于等于40并小于等于60的自然数，例如n取40)的所述主视点和辅助视点进行编码。首先利用HTM8.0软件对当前编码单元相邻的候选空间和时间编码单元位置进行全位置搜索方法以得到当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息，然后利用视差矢量采用率信息删除当前编码单元相邻的候选空间和时间编码单元位置中最少被搜索的位置，将相邻的候选空间和时间编码单元位置进行分组，将其搜索到的视差矢量按照采用率的比例组合为最终视差矢量，最终采用本发明提出的方法对剩余序列的辅助视点相邻视差矢量进行获取，改变了第1个被搜索到的视差矢量即被当作最终视差矢量的准则，改变了原有方法通过删减搜索候选位置减少搜索次数的方式给编码质量带来的不足，能在保持了原有快速算法的效率前提下至少提升编码质量。
具体实施中，在计算机中完成以下程序：
第一步：利用所述的视频编码软件HTM8.0对所述的3D-HEVC国际标准测试视频序列Newspaper_CC、GT fly、Undo dancer，Poznan_Hall2，Poznan_Street，Kendo，Balloons，分别选取前n帧(n为大于等于40并小于等于60的自然数)的所述主视点和辅助视点进行编码，对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
在编码前n帧时，采用对当前编码单元相邻的候选空间和时间编码单元位置进行全位置搜索方法，并同时提取当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息，包括：对相邻的候选时间编码单元位置CTR和BR和对相邻的候选空间编码单元位置A1，B1，B0，A0，B2是否含有视差矢量或者运动补偿预测视差矢量的检测，以及通过对当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量或者运动补偿预测视差矢量信息指向的视差参考帧进行差值平方和(SSD，Sum of Squared Difference)计算，以差值平方和信息进行当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量采用率统计。最后各位置采用率统计结果如表1和表2所示；
第二步：利用上一步得到的当前编码单元相邻的候选空间和时间编码单元位置的视差 矢量采用率的信息为分组搜索提供依据：首先针对相邻的候选时间和空间编码单元位置进行删除操作，删除视差矢量采用率最小的编码单元位置。其中各位置采用率统计如表1和表2所示；
第三步：针对上一步删除之后剩余的相邻的候选空间和时间编码单元位置进行分组，通过第二步得到参考搜索位置包含空间位置的A1，B1，B0，和A0以及时间参考位置的CRT，如图7和图8所示。分组操作针对邻近位置的编码单元，其中B1和B0相邻，A1和A0相邻，则B1和相邻的B0分为一组，A1和相邻的A0分为另一组。其中A1和A0以及B1和B0分组后的结合比例如表3所示；
第四步：设置获取视差矢量标志变量，视差矢量标志变量表征视差矢量是否获取，已获取则为1，未获取则为0值。视差矢量设定为初始的(0，0)，并按照在第一步所述对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
第五步：在时间方向参考帧的相邻视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则检测参考帧对应位置的CTR位置，如果视差矢量可以检测得到，则视差矢量标志变量设为1；
第六步：进行空间位置的视差补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2。空间方向帧内相邻块视差补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到视差补偿预测视差矢量，再继续搜索A0，判断是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将A1中的视差补偿预测视差矢量和A0中的视差补偿预测视差矢量基第三步中分组结合比例表(表3)结合为A1位置视差矢量并采用，标志变量设置为1，若A0未找到视差补偿预测视差矢量，则采用A1中的视差补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历。若A1中未找到视差补偿预测视差矢量，则跳过A0直接检测B1位置的视差补偿预测视差矢量，若B1中找到视差补偿预测视差矢量，则判断B0位置是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将B1中视差补偿预测视差矢量和B0中的视差补偿预测视差矢量以基于第三步中分组结合比例表(表3)结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1。若组2中B1未找到视差补偿预测视差矢量则跳过本步骤；
第七步：进行空间位置的运动补偿预测视差矢量检测，将A1和A0的分组标记为组1， B1和B0的组标记为组2。空间方向帧内相邻块运动补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到运动补偿预测视差矢量，再继续搜索A0，判断是否找到运动补偿预测视差矢量，若运动补偿预测视差矢量找到则将A1中的运动补偿预测视差矢量和A0中的运动补偿预测视差矢量基于第三步中分组结合比例表(表3)结合为A1位置视差矢量并采用，标志变量设置为1，若A0未找到运动补偿预测视差矢量，则采用A1中的运动补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历。若A1中未找到运动补偿预测视差矢量，则跳过A0直接检测B1位置的运动补偿预测视差矢量，若B1中找到运动补偿预测视差矢量，则判断B0位置是否找到运动补偿预测视差矢量，若视运动补偿预测视差矢量找到则将B1中运动补偿预测视差矢量和B0中的运动补偿预测视差矢量以基于第三步中分组结合比例表(表3)结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1。若组2中B1未找到运动补偿预测视差矢量则跳过本步骤；
第八步：判断深度优化标志，若标志位为1采用深度优化操作。深度优化操作(参考文献1中DoNBDV定义)为已有技术。首先将面找的最终视差矢量其指向的纹理块。将最终视差矢量指向的纹理块对应的深度块，搜索深度块4个角的像素位置的深度值选取其最大值。最后将这个最大值转换为深度优化后的视差矢量，转换公式为：
Figure PCTCN2015077944-appb-000015
其中
Figure PCTCN2015077944-appb-000016
为深度优化后的视差矢量，f为摄像机焦距值，l为基线距离，dmax为搜索深度块4个角的像素位置的最大深度值，
Figure PCTCN2015077944-appb-000017
和
Figure PCTCN2015077944-appb-000018
为离摄像机最近和最远的位置坐标(其中f，l和
Figure PCTCN2015077944-appb-000019
都是摄像机参数表中的已有固定参数值)；
本发明改变了第1个被搜索到的视差矢量即被当作最终视差矢量的准则，改变了原有方法通过删减搜索候选位置减少搜索次数的方式给编码质量带来的不足，能在保持了原有快速算法的效率前提下至少提升编码质量，对测试序列对比现有快速算法实验结果如表4所示。

Claims (1)

  1. 一种面向多纹理多深度视频编码中辅助视点编码中相邻视差矢量获取方法，其特征在于，是在计算机中依次按以下步骤仿真实现的：
    步骤(1)，计算机初始化：
    设置：按照通用测试条件对所选定序列的主视点以及当前辅助视点进行编码，编码软件采用3D-HEVC视频编码软件HTM8.0版本作为编码平台；按照通用测试条件规定的纹理图和对应深度图量化参数分为第1组(40，45)，第2组(35，42)，第3组(30，39)，第4组(25，34)，其中括号中前面的数字代表纹理图量化参数，后面为深度图量化参数，共4组；
    以YUV为格式的3D-HEVC国际标准测试视频序列包括：Newspaper_CC、GT fly、Undo dancer，Poznan_Hall2，Poznan_Street，Kendo，Balloons，每个测试序列包含3个纹理序列，以及对应3个深度序列；各个测试序列进行编码时输入编码器的3个视点顺序分别为：4-2-6，5-9-1，5-1-9，6-7-5，4-5-3，3-1-5，3-1-5，在对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
    步骤(2)，利用所述的视频编码软件HTM8.0对所述的3D-HEVC国际标准测试视频序列Newspaper_CC、GT fly、Undo dancer，Poznan_Hall2，Poznan_Street，Kendo，Balloons，分别选取前n帧的所述主视点和辅助视点进行编码，n为大于等于40并小于等于60的自然数，对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；在编码前n帧时，采用对当前编码单元相邻的候选空间和时间编码单元位置进行全位置搜索方法，并同时提取当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息，包括：对相邻的候选时间编码单元位置CTR和BR和对相邻的候选空间编码单元位置A1，B1，B0，A0，B2是否含有视差矢量或者运动补偿预测视差矢量的检测，以及通过对当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量或者运动补偿预测视差矢量信息指向的视差参考帧进行差值平方和计算，以差值平方和信息进行当前编码单元所有相邻的候选空间和时间编码单元位置找到的视差矢量采用率统计；；
    步骤(3)，对所述多纹理多深度视频中除主视点以外的所有辅助视点的编码帧的编码单元按照对当前编码单元相邻的候选空间和时间编码单元位置分组搜索并依采用率的比例组合为最终视差矢量方法进行相邻视差矢量获取，其步骤如 下：
    步骤(3.1)，利用步骤(2)得到的当前编码单元相邻的候选空间和时间编码单元位置的视差矢量采用率的信息为分组搜索提供依据：首先针对相邻的候选时间和空间编码单元位置进行删除操作，删除视差矢量采用率最小的编码单元位置；针对删除之后剩余的相邻的候选空间和时间编码单元位置进行分组，得到参考搜索位置包含空间位置的A1，B1，B0，和A0以及时间参考位置的CRT；分组操作针对邻近位置的编码单元，其中B1和B0相邻，A1和A0相邻，则B1和相邻的B0分为一组，A1和相邻的A0分为另一组；
    步骤(3.2)，利用步骤(3.1)得到的当前编码单元对相邻的候选空间和时间编码单元搜索位置和分组信息进行相邻视差矢量的搜索和最终视差矢量的合成计算操作，其步骤如下：
    步骤(3.2.1)，设置获取视差矢量标志变量，视差矢量标志变量表征视差矢量是否获取，已获取则为1，未获取则为0值；视差矢量设定为初始的(0，0)，并按照在步骤(1)所述对每个视点进行编码时采用纹理图编码完成后再进行深度图编码的顺序进行；
    步骤(3.2.2)，在时间方向参考帧的相邻视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则检测参考帧对应位置的CTR位置，如果视差矢量可以检测得到，则视差矢量标志变量设为1；
    步骤(3.2.3)，进行空间位置的视差补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2；空间方向帧内相邻块视差补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到视差补偿预测视差矢量，再继续搜索A0，判断是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将A1中的视差补偿预测视差矢量和A0中的视差补偿预测视差矢量基于步骤(3.1)中分组结合为A1位置视差矢量并采用，标志变量设置为1，若A0未找到视差补偿预测视差矢量，则采用A1中的视差补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历；若A1中未找到视差补偿预测视差矢量，则跳过A0直接检测B1位置的视差补偿预测视差矢量，若B1中找到视差补偿预测视差矢量，则判断B0位置是否找到视差补偿预测视差矢量，若视差补偿预测视差矢量找到则将B1中视差补偿预测视差矢 量和B0中的视差补偿预测视差矢量以基于步骤(3.1)中分组结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1；若组2中B1未找到视差补偿预测视差矢量则跳过步骤(3.2.3)；
    步骤(3.2.4)，进行空间位置的运动补偿预测视差矢量检测，将A1和A0的分组标记为组1，B1和B0的组标记为组2；空间方向帧内相邻块运动补偿预测视差矢量获取方式为，检测视差矢量标志变量是否为0，为0则判断组1中A1，若A1中找到运动补偿预测视差矢量，再继续搜索A0，判断是否找到运动补偿预测视差矢量，若运动补偿预测视差矢量找到则将A1中的运动补偿预测视差矢量和A0中的运动补偿预测视差矢量基于步骤(3.1)中分组结合为A1位置视差矢量并采用，标志变量设置为1，若A0未找到运动补偿预测视差矢量，则采用A1中的运动补偿预测视差矢量，寻找程序终止跳出，后续位置不再遍历；若A1中未找到运动补偿预测视差矢量，则跳过A0直接检测B1位置的运动补偿预测视差矢量，若B1中找到运动补偿预测视差矢量，则判断B0位置是否找到运动补偿预测视差矢量，若视运动补偿预测视差矢量找到则将B1中运动补偿预测视差矢量和B0中的运动补偿预测视差矢量以基于步骤(3.1)中分组结合为B2位置的视差矢量并采用，视差矢量标志变量设置为1；若组2中B1未找到运动补偿预测视差矢量则跳过步骤(3.2.4)；
    步骤(3.2.5)，判断深度优化标志，若标志位为1采用深度优化操作；首先将面找的最终视差矢量其指向的纹理块；将最终视差矢量指向的纹理块对应的深度块，搜索深度块4个角的像素位置的深度值选取其最大值；最后将这个最大值转换为深度优化后的视差矢量，转换公式为：
    Figure PCTCN2015077944-appb-100001
    其中
    Figure PCTCN2015077944-appb-100002
    为深度优化后的视差矢量，f为摄像机焦距值，l为基线距离，dmax为搜索深度块4个角的像素位置的最大深度值，
    Figure PCTCN2015077944-appb-100003
    和
    Figure PCTCN2015077944-appb-100004
    为离摄像机最近和最远的位置坐标，其中f，l和
    Figure PCTCN2015077944-appb-100005
    都是摄像机参数表中的已有固定参数值。
PCT/CN2015/077944 2015-04-01 2015-04-30 一种面向多纹理多深度视频的相邻视差矢量获取方法 WO2016155070A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/316,009 US9883200B2 (en) 2015-04-01 2015-04-30 Method of acquiring neighboring disparity vectors for multi-texture and multi-depth video

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510152909.3A CN104768019B (zh) 2015-04-01 2015-04-01 一种面向多纹理多深度视频的相邻视差矢量获取方法
CN201510152909.3 2015-04-01

Publications (1)

Publication Number Publication Date
WO2016155070A1 true WO2016155070A1 (zh) 2016-10-06

Family

ID=53649575

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/077944 WO2016155070A1 (zh) 2015-04-01 2015-04-30 一种面向多纹理多深度视频的相邻视差矢量获取方法

Country Status (3)

Country Link
US (1) US9883200B2 (zh)
CN (1) CN104768019B (zh)
WO (1) WO2016155070A1 (zh)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9993715B2 (en) * 2016-01-27 2018-06-12 Cfph, Llc Instructional surface with haptic and optical elements
CN107239268A (zh) 2016-03-29 2017-10-10 阿里巴巴集团控股有限公司 一种业务处理方法、装置和智能终端
CN111726639B (zh) 2016-11-18 2023-05-30 上海兆芯集成电路有限公司 纹理砖压缩及解压缩方法以及使用该方法的装置
WO2021054460A1 (ja) * 2019-09-18 2021-03-25 パナソニック インテレクチュアル プロパティ コーポレーション オブ アメリカ 三次元データ符号化方法、三次元データ復号方法、三次元データ符号化装置、及び三次元データ復号装置
CN110636304B (zh) * 2019-10-23 2021-11-12 威创集团股份有限公司 一种YCbCr444和YCbCr422转换方法
CN111246218B (zh) * 2020-01-16 2023-07-14 郑州轻工业大学 基于jnd模型的cu分割预测和模式决策纹理编码方法
CN114255173A (zh) * 2020-09-24 2022-03-29 苏州科瓴精密机械科技有限公司 粗糙度补偿方法、系统、图像处理设备及可读存储介质
CN112243131B (zh) * 2020-10-31 2022-11-11 西安邮电大学 基于可重构阵列处理器的先进残差预测方法
CN116208756A (zh) * 2023-03-22 2023-06-02 南通大学 一种基于多层级特征融合的深度图快速编码方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1767655A (zh) * 2005-10-18 2006-05-03 宁波大学 一种多视点视频图像视差估计的方法
CN103533361A (zh) * 2013-10-21 2014-01-22 华为技术有限公司 多视点视差矢量的确定方法、编码设备及解码设备
WO2014047351A2 (en) * 2012-09-19 2014-03-27 Qualcomm Incorporated Selection of pictures for disparity vector derivation
CN104365103A (zh) * 2012-06-15 2015-02-18 高通股份有限公司 视频译码中的视差向量选择

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100987765B1 (ko) * 2003-09-30 2010-10-13 삼성전자주식회사 동영상 부호화기에서의 예측 수행 방법 및 장치
CN101061723A (zh) * 2004-11-22 2007-10-24 皇家飞利浦电子股份有限公司 涉及覆盖和无覆盖的运动向量域的投射
RU2480941C2 (ru) * 2011-01-20 2013-04-27 Корпорация "Самсунг Электроникс Ко., Лтд" Способ адаптивного предсказания кадра для кодирования многоракурсной видеопоследовательности
CN103139569B (zh) * 2011-11-23 2016-08-10 华为技术有限公司 多视点视频的编码、解码方法、装置和编解码器
CN102420990B (zh) * 2011-12-15 2013-07-10 北京工业大学 一种面向多视点视频的快速编码方法
US20130271565A1 (en) * 2012-04-16 2013-10-17 Qualcomm Incorporated View synthesis based on asymmetric texture and depth resolutions
US9674523B2 (en) * 2012-11-14 2017-06-06 Advanced Micro Devices, Inc. Methods and apparatus for transcoding digital video
US9497485B2 (en) * 2013-04-12 2016-11-15 Intel Corporation Coding unit size dependent simplified depth coding for 3D video coding
US10009621B2 (en) * 2013-05-31 2018-06-26 Qualcomm Incorporated Advanced depth inter coding based on disparity of depth blocks
US9924182B2 (en) * 2013-07-12 2018-03-20 Samsung Electronics Co., Ltd. Method for predicting disparity vector based on blocks for apparatus and method for inter-layer encoding and decoding video
CN105474640B (zh) * 2013-07-19 2019-03-15 寰发股份有限公司 三维视频编码的摄像机参数发信的方法和装置
WO2015006967A1 (en) * 2013-07-19 2015-01-22 Mediatek Singapore Pte. Ltd. Simplified view synthesis prediction for 3d video coding
US20150063464A1 (en) * 2013-08-30 2015-03-05 Qualcomm Incorporated Lookup table coding
CN104469336B (zh) * 2013-09-25 2017-01-25 中国科学院深圳先进技术研究院 多视点深度视频信号的编码方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1767655A (zh) * 2005-10-18 2006-05-03 宁波大学 一种多视点视频图像视差估计的方法
CN104365103A (zh) * 2012-06-15 2015-02-18 高通股份有限公司 视频译码中的视差向量选择
WO2014047351A2 (en) * 2012-09-19 2014-03-27 Qualcomm Incorporated Selection of pictures for disparity vector derivation
CN103533361A (zh) * 2013-10-21 2014-01-22 华为技术有限公司 多视点视差矢量的确定方法、编码设备及解码设备

Also Published As

Publication number Publication date
CN104768019B (zh) 2017-08-11
US9883200B2 (en) 2018-01-30
US20170094306A1 (en) 2017-03-30
CN104768019A (zh) 2015-07-08

Similar Documents

Publication Publication Date Title
WO2016155070A1 (zh) 一种面向多纹理多深度视频的相邻视差矢量获取方法
CN106134191B (zh) 用于低延迟亮度补偿处理以及基于深度查找表的编码的方法
CN102970529B (zh) 一种基于对象的多视点视频分形编码压缩与解压缩方法
TWI461066B (zh) 彈性調整估算搜尋範圍的移動估算方法及視差估算方法
CN103338370B (zh) 一种多视点深度视频快速编码方法
CN104602028B (zh) 一种立体视频b帧整帧丢失错误隐藏方法
CN101986716A (zh) 一种快速深度视频编码方法
CN100581265C (zh) 一种多视点视频的处理方法
CN103475884B (zh) 面向hbp编码格式的立体视频b帧整帧丢失错误隐藏方法
CN103024381B (zh) 一种基于恰可察觉失真的宏块模式快速选择方法
CN102801995A (zh) 一种基于模板匹配的多视点视频运动和视差矢量预测方法
CN110557646B (zh) 一种智能视点间的编码方法
CN105049866A (zh) 基于绘制失真模型的多视点加深度编码的码率分配方法
CN102413332A (zh) 基于时域增强的视点合成预测多视点视频编码方法
TWI489876B (zh) A Multi - view Video Coding Method That Can Save Decoding Picture Memory Space
CN102316323B (zh) 一种快速的双目立体视频分形压缩与解压缩方法
CN104506871B (zh) 一种基于hevc的3d视频快速编码方法
CN109660811A (zh) 一种快速的hevc帧间编码方法
CN108668135A (zh) 一种基于人眼感知的立体视频b帧错误隐藏方法
CN103595991A (zh) 深度视频编码的像素级预测方法
CN104618714B (zh) 一种立体视频帧重要性评估方法
CN107509074B (zh) 基于压缩感知的自适应3d视频压缩编解码方法
CN102263952B (zh) 一种基于对象的快速双目立体视频分形压缩与解压缩方法
CN102263953B (zh) 一种基于对象的快速多目立体视频分形压缩与解压缩方法
CN109922349B (zh) 基于视差矢量外推的立体视频右视点b帧错误隐藏方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15887038

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 15316009

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 15887038

Country of ref document: EP

Kind code of ref document: A1