Entropy Slice Context Initialization for Parallel Video Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
State-of-the-art video-coding methods like H.264/AVC and TMuC provide higher coding efficiency but at the cost of increased complexity, with entropy decoding often being a bottleneck due to serial processing within slices and reliance on context models that require previous symbol decoding, limiting parallel processing capabilities.
Innovation Solution
The method involves initializing context models for entropy coding at the start of an entropy slice, allowing each entropy slice to maintain its own set of context models and enabling parallel entropy decoding by partitioning a reconstruction slice into multiple entropy slices, where each slice has a self-contained entropy coding state and restricted size to optimize processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If context models are initialized at the start of each slice using a single shared context table, then coding efficiency is improved through context-adaptive binary arithmetic coding, but parallel processing capability deteriorates due to serial dependency within slices
Solution Approach 1:
The patent divides a reconstruction slice into multiple entropy slices, where each entropy slice has its own separate context table for entropy decoding. This segmentation allows different entropy slices to be decoded in parallel without interfering with each other's context models, thereby resolving the contradiction between maintaining coding efficiency through context adaptation and enabling parallel processing.
2Quantity of substance
If a single context table is shared across all entropy slices, then memory usage is reduced, but decoding speed deteriorates due to serial processing requirements
Solution Approach 1:
The patent creates multiple context tables, one for each entropy slice, allowing parallel decoding of different entropy slices. While this increases total memory usage compared to a single shared table, it enables simultaneous processing of multiple slices, thereby improving decoding speed and resolving the contradiction between memory efficiency and processing speed.
3Measurement precision
If context models are updated continuously during decoding, then coding precision is improved through adaptation to local statistics, but processing complexity increases due to serial dependency
Solution Approach 1:
The patent partitions the decoding process into multiple entropy slices, each with its own context table that is updated independently during decoding of that slice. This segmentation allows context adaptation to maintain coding precision within each slice while enabling parallel processing across slices, thereby reducing overall processing complexity and resolving the contradiction between precision and complexity.
4Productivity
If entropy slices are made larger to reduce overhead, then processing efficiency per slice is improved, but parallel processing capability deteriorates due to increased inter-slice dependencies
Solution Approach 1:
The patent creates multiple smaller entropy slices within a reconstruction slice, each with its own context table. This segmentation strategy balances slice size to minimize overhead while maximizing parallel processing capability, as smaller slices have fewer internal dependencies and can be processed more independently in parallel, resolving the contradiction between per-slice efficiency and parallel flexibility.
Data Source
AI summary
Aspects of the present invention are related to systems and methods for context model initialization in entropy encoders and decoders.