Extended Transform Partitions for Video Compression

Abstract

Transform-level partitioning of a prediction residual block is performed to improve compression efficiency of video data. During encoding, a prediction residual block is generated responsive to prediction-level partitioning performed against a video block, a transform block partition type to use is determined based on the prediction residual block, a non-recursive transform-level partitioning is performed against the prediction residual block according to the transform block partition type, and transform blocks generated as a result of the transform-level partitioning are encoded to a bitstream. During decoding, a symbol representative of the transform block partition type used to encode transform blocks is derived from the bitstream, inverse transformed blocks are produced by inverse transforming encoded video data associated with the prediction residual block, and the prediction residual block is reproduced according to the transform block partition type and used to reconstruct the video block, which is output within an output video stream.

What is claimed is: 
     
         1 . A method, comprising:
 performing transform-level partitioning limited to a single, non-recursive partitioning to partition a prediction residual into one or more sub-blocks;   transforming video data of a sub-block of the one or more sub-blocks; and   encoding the transformed video data to a bitstream.   
     
     
         2 . The method of  claim 1 , wherein the single, non-recursive partitioning is according to a transform block partition type indicating a number and arrangement of the one or more sub-blocks. 
     
     
         3 . The method of  claim 2 , comprising:
 encoding an indication of the transform block partition type to the bitstream within a block header, a slice header, a tile header, or a frame header associated with a current block from which the prediction residual is determined.   
     
     
         4 . The method of  claim 2 , comprising:
 determining the transform block partition type for the prediction residual using a machine learning model trained based on one or more of prediction residual data, transform block data, or bitstream data.   
     
     
         5 . The method of  claim 2 , comprising:
 determining the transform block partition type for the prediction residual based on a size of the prediction residual.   
     
     
         6 . The method of  claim 2 , comprising:
 determining the transform block partition type for the prediction residual using a rate-distortion scheme.   
     
     
         7 . The method of  claim 2 , wherein the transform block partition type is determined independent of whether the prediction residual is produced using inter-prediction or intra-prediction. 
     
     
         8 . The method of  claim 2 , wherein the prediction residual corresponds to a first prediction residual sub-block partitioned from a current block during a prediction-level partitioning, and wherein the method comprises:
 performing, using a second transform block partition type, transform-level partitioning limited to a single, non-recursive partitioning to partition a second prediction residual into one or more second sub-blocks, wherein the second prediction residual corresponds to a second prediction residual sub-block partitioned from the current block during the prediction-level partitioning.   
     
     
         9 . The method of  claim 1 , wherein the one or more sub-blocks include one sub-block covering an entirety of the prediction residual. 
     
     
         10 . The method of  claim 1 , wherein the one or more sub-blocks include four sub-blocks arranged in two rows and two columns of the prediction residual. 
     
     
         11 . The method of  claim 1 , wherein the one or more sub-blocks include two sub-blocks arranged next to one another and each covering one half of the prediction residual. 
     
     
         12 . The method of  claim 1 , wherein the one or more sub-blocks include four sub-blocks arranged next to one another and each covering one fourth of the prediction residual. 
     
     
         13 . A method, comprising:
 reconstructing a current block by decoding, according to a transform block partition type derived from a bitstream, transformed video data of one or more sub-blocks encoded based on a transform-level partitioning performance limited to a single, non-recursive transform-level partitioning of a prediction residual; and   outputting the current block within an output video stream for storage or display.   
     
     
         14 . The method of  claim 13 , wherein reconstructing the current block comprises:
 inverse transforming and combining the transformed video data of the one or more sub-blocks to produce at least a portion of the prediction residual; and   reconstructing the current block using the prediction residual.   
     
     
         15 . The method of  claim 14 , wherein the at least the portion of the prediction residual is a first portion of the prediction residual, wherein reconstructing the current block comprises:
 producing a second portion of the prediction residual by inverse transforming and combining second transformed video data of one or more second sub-blocks; and   combining the first portion of the prediction residual and the second portion of the prediction residual.   
     
     
         16 . The method of  claim 13 , wherein reconstructing the current block comprises:
 decoding an indication of the transform block partition type from the bitstream to determine a number and arrangement of the one or more sub-blocks.   
     
     
         17 . A method, comprising:
 decoding an encoded block by inverse transforming video data corresponding to one or more sub-blocks encoded based on a transform-level partitioning performance limited to a single, non-recursive transform-level partitioning of a prediction residual; and   outputting the decoded block within an output video stream for storage or display.   
     
     
         18 . The method of  claim 17 , wherein the inverse transforming is according to a transform block partition type indicating a number and arrangement of the one or more sub-blocks. 
     
     
         19 . The method of  claim 18 , wherein decoding the encoded block comprises:
 inverse transforming and combining the one or more sub-blocks according to the transform block partition type.   
     
     
         20 . The method of  claim 18 , wherein decoding the encoded block comprises:
 entropy decoding a symbol representative of the transform block partition type from a bitstream from which the encoded block is decoded.