Intracoding Transform Matrix Reduction for Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression techniques, such as Mode-Dependent Directional Transform (MDDT), require storage of two transform matrices for each spatial prediction mode, leading to high computational complexity and resource inefficiency, particularly in resource-limited devices.
Innovation Solution
Implementing prediction mode dependent directional transform coding using a single directional transform matrix determined through eigenvalue decomposition on auto-correlated residual blocks, and approximating this matrix as a product of a DCT matrix and a sparse matrix to reduce storage and computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If two transform matrices are stored for each spatial prediction mode in MDDT, then transform coding accuracy is improved, but hardware implementation size and computational complexity increase
Solution Approach 1:
The patent merges the two separate transform matrices (Bi and Ai) into a single combined transform matrix T. This is achieved by defining T = Bi * Ai, where the combined matrix captures the essential transformation properties. The single matrix T is then stored and applied as: Y = T * X, where X is the residual block. This merging reduces storage requirements from 2*N² elements to N² elements, while maintaining transform coding accuracy through the mathematical equivalence of the combined transformation.
Solution Approach 2:
The patent changes the parameter representation from two separate matrices to a single matrix with optimized structure. The single transform matrix T is designed to incorporate the directional properties of both Bi and Ai, allowing it to perform the same function with reduced parameters. This parameter reduction directly decreases hardware implementation size while preserving the transform coding accuracy needed for effective compression.
2Measurement precision
If two transform matrices are stored for each spatial prediction mode in MDDT, then transform coding accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent combines two matrix multiplication operations into one. Instead of computing Y = Bi * (X * Ai) which requires two separate N×N matrix multiplications, the patent pre-computes T = Bi * Ai and then computes Y = T * X. This single matrix multiplication reduces the computational complexity from O(2*N³) to O(N³), halving the number of multiply-accumulate operations required while maintaining the same transform coding accuracy.
3Device complexity
If a single directional transform matrix is used, then storage and computational resources are reduced, but coding efficiency may be compromised
Solution Approach 1:
The patent optimizes the single transform matrix T to maintain coding efficiency by incorporating directional properties. The matrix T is designed with specific structural characteristics that preserve the directional transform capabilities needed for effective compression. By carefully constructing T to capture the essential transformation properties of both Bi and Ai, the patent achieves reduced storage requirements without significant loss in coding efficiency, as the single matrix still performs the necessary directional scaling and rotation operations.
Data Source
AI summary
A method of intra-coding blocks of pixels in a digital video sequence is provided that includes selecting a block transform of a plurality of block transforms according to a spatial prediction mode used in generating a block of pixel residual values from a block of pixels, wherein the block transform is based on a single directional transform matrix predetermined for the spatial prediction mode and is a same size as the block of pixel values, applying the block transform to the block of pixel residual values to generate transform coefficients of the residual pixel values, and entropy coding the generated transform coefficients.