Adaptive LFNST Matrix Selection for Low-Complexity Image Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing demand for high-resolution, high-quality images and videos, particularly in virtual reality and augmented reality, requires a highly efficient image/video compression technique to minimize transmission and storage costs while considering computational complexity.
Innovation Solution
The application of a Low-Frequency Non-Separable Transform (LFNST) in image coding, where the LFNST kernel is set based on intra prediction mode and block size, with matrices derived as 16×16 or 32×64 dimensional matrices depending on block dimensions, to enhance coding performance and reduce complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high-resolution, high-quality image/video compression is applied, then coding performance is improved, but computational complexity increases
Solution Approach 1:
The patent applies parameter changes by dynamically selecting LFNST matrix dimensions (16×16 or 32×64) based on block size and intra prediction mode. This allows the system to adjust computational complexity according to the specific coding conditions, achieving good compression performance for small blocks while maintaining efficiency for larger blocks.
Solution Approach 2:
The patent implements dynamics by making the LFNST matrix selection adaptive rather than fixed. The system dynamically chooses between different matrix dimensions and applies LFNST only when beneficial (based on prediction mode and block size), allowing the compression algorithm to adapt to varying image/video characteristics and coding scenarios.
2Productivity
If LFNST matrix size is increased, then coding efficiency is improved, but processing time increases
Solution Approach 1:
The patent applies local quality by tailoring the LFNST matrix size to the specific local characteristics of each block. Smaller blocks use 16×16 matrices for faster processing, while larger blocks use 32×64 matrices for better compression. This localized adaptation ensures optimal balance between coding efficiency and processing time for each region.
Solution Approach 2:
The patent applies partial action by selectively applying LFNST only when beneficial (based on prediction mode and block size criteria) rather than universally applying it to all blocks. This partial application reduces unnecessary computational overhead while maintaining coding efficiency where LFNST provides the most benefit.
Data Source
AI summary
An image decoding method according to the present document comprises a step of deriving a corrected transform coefficient by applying low-frequency non-separable transform (LFNST) to a transform coefficient, wherein an LFNST set for applying the LFNST may be derived on the basis of an intra-prediction mode applied to the current block, and an LFNST matrix may be derived on the basis of the size of the current block and the LFNST set, the LFNST matrix derived as a 16×16 dimensional matrix if the width or the height of the current block is 4 and the width and the height are both 4 or greater, and the LFNST matrix derived as a 64×32 dimensional matrix if the width or the height of the current block is 8 and the width and the height are both 8 or greater. Therefore, coding performance capable of being brought by the LFNST can be maximized within the implementation complexity permitted in forthcoming standards.