Video Partition Entropy Coding With Adaptive Context Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding systems face inefficiencies in entropy coding of partitioning information, particularly in versatile video coding (VVC), which affects compression efficiency and quality.
Innovation Solution
The proposed methods involve adaptive entropy coding techniques that utilize context models and content-dependent probability values for partitioning decisions, including methods such as utilizing reference picture CU partitioning information, bilateral matching, and different probability models for horizontal and vertical binary trees, to enhance entropy encoding and decoding processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional entropy coding methods are used for partitioning information in VVC, then the coding process is simpler, but compression efficiency is reduced
Solution Approach 1:
The patent implements dynamic context model adaptation where probability values for entropy coding are adjusted based on video content characteristics and partitioning patterns. The context models are updated adaptively during encoding based on observed statistics, allowing the system to optimize compression efficiency dynamically rather than using fixed probability values. This resolves the contradiction by making the coding process adaptive to content while maintaining manageable complexity through structured update mechanisms.
Solution Approach 2:
The patent changes the probability parameters used in entropy coding based on video content type and partitioning decisions. Different context models with distinct probability values are selected and applied according to the specific coding situation, such as intra vs inter prediction modes, different block sizes, and content characteristics. This parameter adaptation improves compression efficiency without requiring complete redesign of the entropy coding framework.
2Measurement precision
If content-dependent probability values are used for partitioning decisions, then coding accuracy improves, but processing complexity increases
Solution Approach 1:
The patent segments the entropy coding process into multiple context models, each specialized for different partitioning scenarios. Instead of using a single complex probability model, the system divides the coding space into segments (different context models) based on prediction mode, block size, and content type. Each segment has its own probability values optimized for that specific scenario, improving accuracy while keeping individual model complexities manageable.
Solution Approach 2:
The patent applies local quality by using different probability values and context models tailored to specific local coding conditions. Each partitioning decision uses context information locally relevant to that region and coding situation, rather than applying a global uniform model. This localized adaptation improves decision accuracy while the modular structure keeps processing complexity organized and manageable.
3Productivity
If multiple context models are used for different partitioning scenarios, then coding efficiency improves, but implementation complexity increases
Solution Approach 1:
The patent creates context models that serve multiple functions across different partitioning scenarios. The same context model structure is reused for both intra and inter prediction modes, different block sizes, and various content types, with parameters adapted rather than completely different models. This multi-functionality approach improves compression efficiency across diverse scenarios while avoiding the complexity of maintaining entirely separate models for each case.
Solution Approach 2:
The patent implements a hierarchical structure where general context models contain specialized sub-models or parameters for specific scenarios. The context model system is organized in nested levels, with overarching models that can be specialized for particular partitioning situations. This nesting allows efficient management of multiple context models by organizing them in a structured hierarchy, improving compression while keeping implementation complexity manageable through systematic organization.
Data Source
AI summary
Methods of entropy coding of block partitioning with initial context model probability values indicated by high-level syntaxes or with different context models for HBT (Horizontal Binary Tree) and VBT (Vertical Binary Tree). For one method, one or more coded bits, including encoded data for information related to a partitioning tree of the current picture area, are signalled or parsed from a bitstream. Entropy coding is applied to the coded bits using context formation, including one or more initial context model probability values derived according to a video content type of the current picture, to recover the partitioning tree information. For another method, entropy coding is applied to the coded bits by using one or more context models to recover the first information related to the partitioning tree and the context models are different for HBT and VBT applied to a non-square block.