Image Coding Transform Selection Using LFNST and MTS
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing demand for high-resolution and high-quality images/videos, such as 4K and 8K ultra high definition, has led to higher transmission and storage costs due to increased bit amounts. Additionally, the need for efficient compression techniques is heightened by the growing interest in immersive media like virtual and augmented reality.
Innovation Solution
The proposed solution involves an image coding method and apparatus that utilize Low-Frequency Non-Separable Transform (LFNST) and Multiple Transform Selection (MTS) to enhance image coding efficiency. This method includes deriving residual samples by applying LFNST or MTS to transform coefficients and generating a reconstructed picture based on these residual samples. The LFNST and MTS indices are signaled to indicate the respective kernels used, with the MTS index being parsed based on the color index of the current block and the LFNST index being parsed regardless of the color index.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high-resolution and high-quality images/videos (4K, 8K UHD) are transmitted or stored, then image quality is improved, but transmission cost and storage cost increase due to increased bit amount
Solution Approach 1:
The patent applies Low-Frequency Non-Separable Transform (LFNST) and Multiple Transform Selection (MTS) to segment the transform process into different frequency components and transform types. By dividing the transform coefficients into low-frequency and high-frequency regions and applying different transform strategies, the patent achieves better compression efficiency for high-resolution images while reducing the bit amount required for transmission and storage.
Solution Approach 2:
The patent changes transform parameters by selecting different transform kernels (DCT, DST, DRT) based on the characteristics of the current block. The MTS index signaling mechanism allows dynamic adjustment of transform parameters, enabling optimal compression for different image regions and frequency components, thereby reducing overall bit amount while maintaining image quality.
2Productivity
If conventional transform methods are used for image coding, then device complexity is reduced, but image coding efficiency is insufficient for high-resolution images
Solution Approach 1:
The patent introduces dynamic transform selection through MTS, where the transform type is selected based on the characteristics of the current block (e.g., prediction mode, block size). This dynamic adaptation allows the system to optimize coding efficiency for different image regions without requiring complex processing for all blocks uniformly. The LFNST is dynamically applied to low-frequency components, providing enhanced compression where needed.
Solution Approach 2:
The patent applies different transform strategies to different regions of the transform coefficient block. LFNST is specifically applied to low-frequency components, while MTS selects appropriate transforms for high-frequency components based on local characteristics. This localized approach improves overall coding efficiency without uniformly increasing complexity across the entire image processing system.
3Measurement precision
If multiple transform indices (LFNST index and MTS index) are signaled, then transform precision is improved, but bit amount increases
Solution Approach 1:
The patent applies LFNST selectively only to low-frequency components rather than the entire transform coefficient block. The MTS index is signaled only when beneficial, and the LFNST index is parsed conditionally based on the current block characteristics. This partial application approach provides precise transform kernel selection where needed while avoiding unnecessary bit overhead in regions where simple transforms suffice.
Solution Approach 2:
Instead of signaling transform indices for all blocks uniformly, the patent inverts the approach by using default transform behavior for most cases and only signaling indices when deviations from the default are beneficial. The LFNST index parsing is conditioned on specific block characteristics, reducing the frequency of index signaling while maintaining transform precision where it provides the most value.
Data Source
AI summary
An image decoding method, according to the present document, comprises the steps of: deriving residual samples by applying at least one among LFNST or MTS to a transform coefficient; and generating a reconstructed picture on the basis of the residual samples, wherein the LFNST is carried out on the basis of a LFNST index indicating a LFNST kernel, the MTS is carried out on the basis of a MTS index indicating a MTS kernel, the MTS index is parsed on the basis that the color index of the current block is a luma component, and the LFNST index may be parsed regardless of the color index of the current block.