MIP Intra Prediction Filtering for Flexible LFNST Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The application of low-frequency non-separable transform (LFNST) technology in matrix-based intra prediction (MIP) mode for versatile video coding (VVC) adversely affects coding efficiency.
Innovation Solution
Determine a prediction parameter for a current block, including a prediction mode parameter, and use matrix-based intra prediction (MIP) to derive an intra prediction value, followed by low-frequency non-separable transform (LFNST) on the prediction residual, with fixed shifting parameters, and encode the LFNST parameter into a bitstream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If LFNST technology is applied to MIP mode prediction, then transform capability is enhanced, but coding efficiency deteriorates
Solution Approach 1:
The patent extracts the LFNST application from the general transform process and creates a separate, dedicated transform unit specifically for low-frequency components. This extraction allows the LFNST to be applied selectively only where beneficial, rather than universally to all transform coefficients, thereby maintaining prediction precision while avoiding the coding efficiency penalty of universal application.
Solution Approach 2:
The patent applies different transform strategies to different frequency regions. Low-frequency coefficients undergo LFNST processing while high-frequency coefficients use conventional transforms. This local differentiation ensures that the enhanced prediction precision from LFNST is applied only to the frequency region where it provides benefit, avoiding unnecessary complexity and bit overhead in regions where it would not improve coding efficiency.
2Manufacturing precision
If LFNST is applied to prediction residual, then transform effectiveness is improved, but computational complexity increases
Solution Approach 1:
The patent segments the transform coefficients into low-frequency and high-frequency portions, applying LFNST only to the low-frequency segment. This segmentation reduces the number of coefficients requiring LFNST processing, thereby maintaining transform effectiveness for the most important frequency components while reducing overall computational complexity compared to applying LFNST to all coefficients.
Solution Approach 2:
The patent applies LFNST partially rather than completely - specifically to low-frequency coefficients where it provides the most benefit. This partial action achieves sufficient transform effectiveness for the critical low-frequency components without the excessive computational burden of applying LFNST to all transform coefficients, including those in high-frequency regions where its benefit would be minimal.
3Adaptability or versatility
If LFNST parameter is encoded into bitstream, then transform flexibility is enhanced, but bitstream overhead increases
Solution Approach 1:
The patent performs preliminary determination of whether LFNST should be applied based on block characteristics before encoding. This preliminary action allows the encoder to decide on LFNST application in advance, enabling flexible adaptability to different block types while avoiding unnecessary LFNST parameter encoding in cases where LFNST would not provide benefit, thereby reducing bitstream overhead.
Solution Approach 2:
The patent changes the transform parameter (whether to apply LFNST) based on block characteristics such as prediction mode or block size. This parameter change approach provides transform flexibility by adapting the transform strategy to different block types, while the conditional nature of the parameter change ensures that LFNST parameters are only encoded when actually needed, minimizing bitstream overhead.
Data Source
AI summary
An encoding method, a decoding method, and a decoder are provided. The method includes the following. A prediction parameter of a current block is determined, where the prediction parameter includes a prediction mode parameter. Neighbouring sample values of the current block are obtained and matrix-based intra prediction (MIP) input sample values of the current block are determined according to the neighbouring sample values of the current block, when the prediction mode parameter indicates that an MIP mode is used to determine an intra prediction value of a colour component of the current block. An MIP prediction value of the colour component of the current block is determined according to the MIP input sample values, an MIP weighting matrix, a shifting offset parameter, and a shifting number parameter. The intra prediction value of the colour component of the current block is determined by filtering the MIP prediction value.