Significance Map Partition Selection for Adaptive Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding standards like H.264/AVC and HEVC inefficiently use context models for entropy encoding of significance maps, leading to suboptimal computational speed and compression efficiency due to uniformly distributed contexts that are not adaptive to the actual data distribution in significance maps.
Innovation Solution
Implement non-uniform partitioning of significance maps into parts with context-adaptive encoding and decoding, using predetermined partition sets and dynamic selection processes to balance accuracy and adaptivity, optimizing context usage based on TU size, texture type, and QP value.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If uniform context distribution is used for encoding significance maps, then the encoding process is simple, but compression efficiency deteriorates
Solution Approach 1:
The patent applies local quality by assigning different context models to different spatial regions of the transform unit. Specifically, the significance map is divided into multiple regions (e.g., top-left, top-right, bottom-left, bottom-right quadrants), and each region uses a separate context model adapted to its local characteristics. This allows the encoding to adapt to local statistical properties of transform coefficients, improving compression efficiency while maintaining manageable complexity through systematic regional division.
2Adaptability or versatility
If the number of contexts is increased for larger transform units, then adaptivity improves, but device complexity increases
Solution Approach 1:
The patent segments the transform unit into multiple smaller regions, each with its own context model. For example, an 8x8 transform unit is divided into four 4x4 regions, and a 16x16 transform unit is divided into four 8x8 regions. This segmentation approach allows adaptivity to be improved for larger transform units by providing region-specific context models, while the total number of contexts remains manageable because each region uses a limited number of context models rather than having a unique context for every coefficient position.
3Manufacturing precision
If more contexts are tracked for significance map encoding, then encoding precision improves, but computational speed decreases
Solution Approach 1:
The patent merges multiple coefficient positions that share similar statistical characteristics into the same context model. Instead of maintaining separate context models for every individual coefficient position, the patent groups positions by their spatial relationships and statistical properties, using a single context model for each group. This merging approach maintains encoding precision by capturing the essential statistical variations through carefully designed context groups, while significantly reducing the computational overhead of tracking and updating numerous separate context models.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Methods of encoding and decoding for video data are describe in which significance maps are encoded and decoded using non-spatially-uniform partitioning of the map into parts, wherein the bit positions within each part are associated with a given context. Example partition sets and processes for selecting from amongst predetermined partition sets and communicating the selection to the decoder are described.