Reduced Secondary Transform Context Modeling for Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards face challenges in efficiently compressing digital video due to high bandwidth demands and sub-optimal transform matrix usage across different video components and prediction modes, leading to increased complexity and parsing delays.
Innovation Solution
The proposed method involves using a reduced secondary transform (RST) with adaptable transform matrices and zero-out regions, applied selectively based on intra prediction modes and color components, to optimize transform operations and reduce unnecessary signaling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a full-size transform matrix (N×N) is used for secondary transform, then transform completeness is maintained, but computational complexity and processing time increase
Solution Approach 1:
The transform block is divided into multiple sub-blocks, and the secondary transform is applied selectively to specific sub-blocks based on prediction mode and coefficient distribution patterns, rather than applying the full transform to the entire block
Solution Approach 2:
A reduced-size transform matrix is used instead of the full N×N matrix, applying transform only to the most significant regions of the transform coefficients where energy is concentrated, achieving adequate compression with reduced computational load
2Reliability
If secondary transform is applied uniformly to all video blocks, then compression consistency is improved, but processing overhead and complexity increase
Solution Approach 1:
Different transform strategies are applied to different regions of the video block based on local characteristics such as intra prediction mode and coefficient distribution patterns, optimizing compression for each region's specific properties
Solution Approach 2:
The secondary transform application is made adaptive and conditional rather than static and uniform, with the transform being selectively applied or skipped based on dynamically determined criteria including prediction mode and coefficient significance patterns
3Productivity
If reduced transform matrices are used, then processing speed and efficiency improve, but transform accuracy and compression performance may deteriorate
Solution Approach 1:
The reduced transform focuses computational resources on specific sub-regions and coefficient patterns where they provide the most benefit, maintaining accuracy for significant components while reducing processing for less important regions
Solution Approach 2:
The transform matrix size and application scope are dynamically adjusted based on block characteristics, using smaller matrices when appropriate to maintain processing efficiency while preserving necessary transform accuracy for achieving target compression performance
Data Source
AI summary
A video processing method is described. The method includes performing a conversion between a video region of a video and a coded representation of the video. The performing of the conversion includes configuring, based on a partition type of the video region, a context model for coding a first bin. The first bin and a second bin are included in a bin string corresponding to an index of a secondary transform tool. The index indicates an applicability of the secondary transform tool and/or a kernel information of the secondary transform tool. The secondary transform tool includes applying, during encoding, a forward secondary transform to an output of a forward primary transform applied to a residual of a video block prior to quantization, or applying, during decoding, an inverse secondary transform to an output of dequantization to the video block before applying an inverse primary transform.