Image Decoding with Mapped Non-Separable Transform Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies face challenges in efficiently encoding and decoding high-resolution and high-quality images due to limitations in transform methods, particularly in handling non-separable transforms and transform kernels.
Innovation Solution
The method employs a non-separable primary transform kernel of reduced dimension, determined based on encoding parameters, and utilizes mapping tables to define transform sets for various block sizes and intra prediction modes, enabling efficient encoding and decoding of images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a non-separable transform is applied to improve transform performance, then encoding efficiency is improved, but device complexity increases
Solution Approach 1:
The patent applies different transform types (separable or non-separable) to different blocks based on their characteristics. Specifically, non-separable transforms are applied to blocks with aspect ratios of 8:1, 1:8, 16:1, or 1:16, while separable transforms are used for other blocks. This localized application optimizes encoding efficiency for specific block types without unnecessarily increasing complexity for all blocks.
Solution Approach 2:
The patent dynamically selects between separable and non-separable transforms based on block aspect ratio. The transform type is not fixed but adapts to the specific characteristics of each block, allowing the system to optimize performance for each local condition while maintaining overall system flexibility.
2Adaptability or versatility
If multiple transform sets are supported for different block sizes and prediction modes, then adaptability is improved, but device complexity increases
Solution Approach 1:
The patent segments the transform selection process into distinct categories based on block aspect ratio. By dividing blocks into specific groups (8:1, 1:8, 16:1, 1:16 ratios versus other ratios), the system can manage multiple transform sets in an organized manner, reducing the complexity of selecting and applying the appropriate transform for each block type.
3Manufacturing precision
If a reduced dimension non-separable transform kernel is used, then manufacturing precision is improved, but loss of information may increase
Solution Approach 1:
The patent changes the parameter of transform kernel dimensionality by using reduced dimension non-separable transform kernels for specific block aspect ratios. This parameter change optimizes the balance between computational precision and information retention, adapting the kernel dimensionality to the specific requirements of different block types to minimize information loss while maintaining manufacturing precision.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image decoding method and device according to the present disclosure can acquire residual information from a bitstream, derive transform coefficients of a current block on the basis of the residual information, derive residual samples of the current block by performing at least one of inverse quantization or inverse transform on the transform coefficients of the current block, and restore the current block on the basis of the residual samples of the current block. Here, the inverse transform can be performed on the basis of at least one of non-separable transform or separable transform. A non-separable transform set for the non-separable transform can be determined on the basis of a mapping table defining mapping relationships between pre-defined intraprediction modes and non-separable transform sets.