Non-Separable Transform Selection for Block-Based Image Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies face challenges in efficiently handling high-resolution and high-quality images, particularly in terms of transform performance and encoding efficiency.
Innovation Solution
The use of a non-separable primary transform (NSPT) and a non-separable primary transform kernel of reduced dimension, applied based on encoding parameters and block sizes, to derive and encode transform coefficients.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a non-separable primary transform is applied to all block sizes, then transform performance improves, but device complexity and computational load increase
Solution Approach 1:
The transform block sizes are segmented into two distinct groups: a first group (e.g., 4x4, 4x8, 8x4) where non-separable primary transform is applied, and a second group (e.g., 8x8, 16x16) where separable primary transform is used. This segmentation allows the system to apply the computationally intensive non-separable transform only where it provides the most benefit (smaller blocks), while using the more efficient separable transform for larger blocks, thus resolving the contradiction between transform performance and computational complexity.
Solution Approach 2:
Different transform methods are applied to different regions (block sizes) based on their specific characteristics. Small blocks receive the high-performance non-separable transform, while large blocks use the computationally lighter separable transform. This local differentiation optimizes overall system performance by matching the transform method to the local requirements of each block size category.
2Productivity
If non-separable primary transform is used for all blocks, then encoding efficiency improves, but loss of time increases
Solution Approach 1:
By segmenting block sizes into two groups with different transform methods, the system achieves encoding efficiency improvements for small blocks (first group) while avoiding the excessive processing time that would result from applying non-separable transform to all blocks including large ones (second group). This selective application optimizes the time-efficiency tradeoff.
Solution Approach 2:
Instead of applying the full non-separable transform to all blocks (excessive action), the system applies it partially only to the first group of block sizes where it provides the most benefit. This partial application avoids the excessive processing time while still capturing the essential efficiency improvements where they are most needed.
3Area of stationary object
If transform block size is increased, then area coverage improves, but transform performance deteriorates
Solution Approach 1:
The solution segments the transform block size range into two groups: small blocks (first group) where non-separable transform maintains high performance, and large blocks (second group) where separable transform is used to avoid performance deterioration. This segmentation acknowledges that transform performance characteristics differ by block size and applies appropriate methods accordingly.
Solution Approach 2:
The system changes the transform method parameter based on the block size parameter. For small block sizes, non-separable transform is used to maintain performance, while for large block sizes, separable transform is selected to prevent performance deterioration. This parameter adaptation resolves the contradiction between area coverage and transform performance.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image decoding method and apparatus according to the present disclosure may comprise: obtaining residual information from a bitstream; deriving transform coefficients of a current block on the basis of the residual information; performing at least one of inverse quantization or inverse transform on the transform coefficients of the current block and deriving residual samples of the current block; and reconstructing the current block on the basis of the residual samples of the current block. In this case, the inverse transform is performed on the basis of a non-separable primary transform, and the non-separable primary transform may be applied on the basis of at least one of the size of a current block, a tree type, and a component type.