Non-separable Transform for GPM Inter-prediction Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding schemes, such as H.264/AVC and HEVC, do not effectively apply non-separable transforms to inter-prediction blocks, which limits coding efficiency due to the presence of many diagonal directions in these blocks.
Innovation Solution
A video decoding and coding apparatus that includes a prediction unit using a weight parameter derived from an angle variable and a non-separable transform processing unit. This apparatus performs a non-separable transform in the Geometric Partition Mode (GPM) for inter-prediction blocks, deriving the necessary transformation matrices from split information in the GPM.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If non-separable transform is applied only to intra-prediction blocks, then intra-prediction coding efficiency is improved, but inter-prediction coding efficiency remains limited
Solution Approach 1:
The non-separable transform is extended to work with both intra-prediction blocks and inter-prediction blocks (specifically GPM mode blocks). The transform processing unit is designed to determine whether to apply non-separable transform based on the prediction mode, enabling the same transform mechanism to serve multiple prediction types and improve overall coding efficiency across different prediction scenarios.
Solution Approach 2:
The patent applies non-separable transform selectively based on local characteristics of the block. For inter-prediction blocks in GPM mode, the transform is applied when the block exhibits directional characteristics suitable for non-separable transform. This selective application ensures that the transform is used where it provides benefit (concentrating coefficients in low-frequency regions) while avoiding unnecessary processing elsewhere.
2Manufacturing precision
If traditional separable transform is used for inter-prediction blocks, then processing complexity is reduced, but coefficient concentration in low-frequency regions is insufficient
Solution Approach 1:
The patent changes the transform parameters by selecting different transform types (separable vs. non-separable) based on the prediction mode and block characteristics. For GPM mode inter-prediction blocks, non-separable transform is selected to achieve better coefficient concentration. The decision is based on parameters such as prediction mode type and block size, allowing the system to adapt the transform parameters to the specific processing context.
3Productivity
If non-separable transform is applied to all blocks, then coding efficiency is maximized, but processing overhead and complexity increase
Solution Approach 1:
Instead of applying non-separable transform to all blocks, the patent applies it partially - specifically to inter-prediction blocks in GPM mode and intra-prediction blocks where it provides benefit. The transform processing unit determines on a per-block basis whether non-separable transform should be applied, avoiding unnecessary processing overhead while maintaining coding efficiency where it matters most.
Data Source
AI summary
Video coding and decoding apparatuses (11 and 31) capable of improving coding efficiency are provided. The video decoding apparatus (31) according to an aspect of the present invention is a video decoding apparatus (31) including a prediction unit (30377) that derives a prediction image using a weight parameter derived from an angle variable, and a non-separable transform processing unit (31121) that performs a non-separable transform, in which the non-separable transform processing unit (31121) derives a set number (lfnstTrSetId) of the non-separable transform from split information (merge_gpm_partition_idx) of the GPM in a case that an inter-prediction mode is a GPM mode.