Non-separable Transform Matrix Index Parsing for Video Coding Throughput
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for performing non-separable transforms in video coding are complex and inefficient, particularly in computing primary transform coefficients, which affects entropy coding throughput due to high computational complexity and large storage requirements for transform matrices.
Innovation Solution
A method that parses transform coefficients to identify non-zero coefficients in a top-left preset region, determines a transform matrix based on an index value, and performs matrix operations to reduce the number of transform matrices used, thereby increasing coding throughput by reducing computational and storage demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-separable transform is performed on primary transform coefficients using current methods, then transform accuracy is improved, but computational complexity increases and coding throughput decreases
Solution Approach 1:
The transform coefficient matrix is divided into multiple sub-matrices, and the non-separable transform is selectively applied only to specific sub-matrices based on prediction mode and coefficient characteristics. This segmentation approach maintains transform accuracy where needed while reducing overall computational complexity and improving coding throughput.
Solution Approach 2:
Different transform strategies are applied to different regions of the coefficient matrix based on local characteristics. Specifically, non-separable transform is applied only to sub-matrices containing significant coefficients or those corresponding to specific prediction modes, while other regions use simpler transform methods, thereby optimizing the balance between accuracy and throughput.
2Adaptability or versatility
If multiple transform matrices are stored for non-separable transform, then transform flexibility is improved, but storage requirements increase
Solution Approach 1:
A small set of universal transform matrices is designed to serve multiple prediction modes and coefficient patterns. These matrices are constructed to capture the essential non-separable transform characteristics across different scenarios, reducing the total number of matrices needed while maintaining transform flexibility and adaptability.
Solution Approach 2:
Instead of storing multiple fixed transform matrices, the patent uses parameter-based selection where a small number of matrices are combined or modified through parameters to adapt to different transform requirements. This approach reduces storage requirements while preserving the ability to handle various transform scenarios.
3Reliability
If non-separable transform index parsing is performed across the entire coefficient matrix, then transform completeness is improved, but parsing complexity increases
Solution Approach 1:
The coefficient matrix is segmented into sub-matrices, and index parsing is performed only on relevant sub-matrices based on prediction mode and coefficient significance. This selective parsing approach maintains transform completeness for important regions while significantly reducing parsing complexity and processing overhead.
Solution Approach 2:
The patent extracts and processes only the essential transform index information from the coefficient matrix, focusing on sub-matrices that contribute most to reconstruction quality. By taking out and processing only critical index data, the method maintains transform completeness while reducing parsing complexity.
Data Source
AI summary
A method includes parsing a transform coefficient of a transform block in a current coding unit to obtain a first transform coefficient matrix. The method also includes obtaining a quantity K of non-zero transform coefficients in a top-left preset region of the first transform coefficient matrix. The method also includes parsing an index value based on the quantity K being greater than a threshold. The method also includes determining a transform matrix based on the index value. The method also includes multiplying N transform coefficients of the first transform coefficient matrix by the transform matrix to obtain M transform coefficients. The method also includes updating the first transform coefficient matrix by using the M transform coefficients to obtain a second transform coefficient matrix. The method also includes performing an inverse transform on the second transform coefficient matrix to obtain residual samples of the current coding unit.