Residual Tree Structure for Video Transform Unit Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression systems, such as HEVC, face inefficiencies in representing motion information due to the need for complex quadtree and residual quadtree data structures that require encoding and transmitting extensive overhead data, especially when partitioning into smaller transform units.
Innovation Solution
A method and apparatus that utilize a residual tree structure allowing binary partitioning of prediction units into two transform units, reducing the complexity and overhead by determining a suitable transform unit partition based on the desired level of partitioning, which can be signaled between encoder and decoder.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a residual quadtree structure with four-branch partitioning is used to represent transform unit partitioning, then the partitioning can describe detailed motion information, but the data overhead for encoding and transmission increases significantly
Solution Approach 1:
The patent segments the residual quadtree structure by separating it into two independent binary trees: one for horizontal partitioning and one for vertical partitioning. This segmentation reduces the complexity from a four-branch node structure to two separate two-branch structures, thereby reducing the number of flags and bits required to encode the partitioning information while maintaining the ability to represent detailed motion information through combined horizontal and vertical splits.
2Productivity
If multiple transform sizes are used in a macroblock, then the compression efficiency improves by adapting to different block characteristics, but a macroblock level parameter is required to indicate transform size which increases overhead
Solution Approach 1:
The patent implements dynamic transform unit partitioning where the transform block size is adaptively determined based on the prediction block type and content characteristics. Instead of using fixed transform sizes or requiring macroblock-level parameters, the system dynamically selects transform sizes (e.g., 4x4, 8x8, 16x16) at the transform unit level, allowing compression efficiency to be optimized locally without increasing overall parameter overhead.
3Adaptability or versatility
If the LCU is partitioned into multiple CUs using quadtree representation, then the coding can be adapted to different regions, but the binary data representation requires encoding and transmission as overhead
Solution Approach 1:
The patent segments the LCU partitioning representation by separating horizontal and vertical splits into distinct binary trees. This allows the coding system to adapt to different regions with finer granularity while reducing the overhead compared to a traditional quadtree. The segmented approach enables independent optimization of horizontal and vertical partitioning patterns without requiring complete quadtree flag sequences to be transmitted.
Data Source
Figure 1~2
Figure 3~5
Figure 6~7
AI summary
In one embodiment, a method includes receives a prediction unit (PU) for a coding unit (CU) of video content. The PU is partitionable into a plurality of PU partition types. The method determines a PU partition type for the PU and a residual tree structure based on the PU partition type for partitioning of the CU into transform units (TUs). The residual tree includes a binary partition of a node into two. A TU partition for the PU partition type is determined based on the residual tree structure and a desired level of partitioning in the residual tree structure. The method then uses the TU partition in a transform operation.