Parallelizing Video Context Selection via Transform Unit Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards face challenges in parallelizing context selection due to dependencies between neighboring binary symbols, which hinders efficient processing and increases computational complexity during decoding.
Innovation Solution
The method involves determining the scan type and analyzing neighboring positions within a transform unit to remove dependencies, enabling parallel processing of at least two bins by modifying the context selection process to reduce reliance on immediate neighbors, particularly at the edges of the transform unit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If sequential context selection is used to maintain coding efficiency, then coding efficiency is improved, but processing speed and parallelization capability deteriorate
Solution Approach 1:
The transform unit is divided into multiple parallel processing regions (e.g., four quadrants) that can be processed simultaneously. Each region has its own context selection logic that operates independently, allowing parallel decoding while maintaining coding efficiency through localized context modeling.
Solution Approach 2:
Context models are pre-initialized and prepared before parallel processing begins. The context selection logic is set up in advance with default states, allowing multiple processing threads to start simultaneously without waiting for sequential context establishment, thus improving processing speed while maintaining coding efficiency.
2Loss of information
If dependencies between neighboring binary symbols are maintained for accurate context selection, then coding efficiency is improved, but parallelization capability deteriorates
Solution Approach 1:
Different regions of the transform unit are assigned different dependency rules. Edge regions maintain stronger dependencies on neighboring symbols for accurate context selection, while interior regions use simplified context models that enable parallel processing. This local differentiation maintains overall coding efficiency while reducing computational complexity for parallelization.
Solution Approach 2:
Instead of removing all dependencies between neighboring symbols, the patent applies partial dependency removal only where necessary to enable parallelization. Full context modeling is maintained for critical regions, while simplified modeling is applied to regions where parallel processing benefits outweigh the slight loss in context accuracy, achieving a balance between coding efficiency and parallelization capability.
Data Source
AI summary
A method and apparatus for parallel processing of at least two bins relating to at least one of a video and an image. The method includes determining scan type of at least a portion of the at least one of video and an image, analyzing neighboring position of a bin, removing dependencies of context selection based on the scan type and position of location being encoded in a transform, and performing parallel processing of that least two bins.