Adaptive LFNST Kernel Selection for Image Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image/video coding technologies face challenges in efficiently compressing and transmitting high-resolution, high-quality images/videos, especially with the increasing demand for VR/AR content and immersive media, leading to increased costs and complexity in storage and transmission.
Innovation Solution
The implementation of a low-frequency non-separable transform (LFNST) method in image coding systems, where modified transform coefficients are derived using specific transform kernels based on block sizes and intra prediction modes to improve coding efficiency and minimize computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution, high-quality image/video data is transmitted or stored, then image quality and resolution are improved, but transmission and storage costs increase
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting transform kernel sizes based on block dimensions. For 4x4 blocks, a 16x16 transform kernel is used, while for 8x8 blocks, a 48x16 transform kernel is used. This adaptive parameter selection optimizes the balance between compression efficiency and computational complexity, enabling effective compression of high-resolution content without requiring proportional increases in transmission and storage resources
2Productivity
If LFNST with larger transform kernels is applied, then coding efficiency is improved, but computational complexity increases
Solution Approach 1:
The patent implements local quality by applying different transform kernel sizes to different block regions based on their dimensions. Smaller blocks (4x4) use smaller transform kernels (16x16), while larger blocks (8x8) use larger transform kernels (48x16). This localized adaptation ensures that computational complexity is proportional to the actual coding needs of each block, maximizing coding efficiency where needed while minimizing complexity where simple blocks are present
3Productivity
If transform kernels are derived based on block sizes, then coding performance is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-defining and storing multiple transform kernels in lookup tables during system initialization or configuration. The kernels for 4x4 blocks (16x16) and 8x8 blocks (48x16) are prepared in advance and stored in memory. During actual coding, the system simply retrieves the appropriate pre-computed kernel from memory based on block dimensions, avoiding the need for real-time kernel derivation and significantly reducing computational complexity while maintaining high coding performance
Data Source
AI summary
An image decoding method according to this document comprises a step for deriving, for transform coefficients, modified transform coefficients on the basis of inverse secondary transformation, wherein the step for deriving the modified transform coefficients comprises a step for deriving a transform kernel to be applied to the inverse secondary transformation, the transform kernel may be derived as a 16×16 matrix on the basis of horizontal and vertical lengths of a target block being both greater than or equal to 4 and the horizontal or vertical length being 4, and the transform kernel may be derived as a 48×16 matrix on the basis of the horizontal and vertical lengths of the target block being both greater than or equal to 8 and the horizontal or vertical length being 8.