Adaptive Separable Graph-Based Transform for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression techniques, such as separable DCT, assume isotropic statistical properties for residual signals, which is not accurate for all video content, leading to inefficiencies in compression and increased computational complexity.
Innovation Solution
A generalized separable graph-based transform is applied, combining two 1D transforms and selecting line graphs based on edge weights and self-loops, allowing for adaptive processing in both column and row directions of coding blocks, enabling flexible and efficient video signal processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If separable DCT is used for encoding residual signals, then the transform can be efficiently implemented, but it assumes isotropic statistical properties which does not match actual residual block characteristics, leading to reduced coding efficiency
Solution Approach 1:
The patent makes the transform dynamically adaptive by selecting different graph-based transform kernels (DKT, DGST, DRDST) based on the statistical properties of residual blocks. The transform type is chosen adaptively according to prediction mode and residual characteristics, allowing the system to optimize for each block rather than using a fixed transform, thus resolving the contradiction between implementation efficiency and adaptability.
Solution Approach 2:
The patent changes the transform parameters by using different graph-based transform kernels with varying characteristics (DCT for cosine-like signals, DST for sine-like signals, RDST for reverse sine-like signals). By changing the transform kernel parameter based on residual block statistics, the system achieves better adaptation while maintaining computational efficiency through the separable structure.
2Productivity
If different transform types are applied to adapt to statistical properties of residual blocks, then coding efficiency is improved, but calculation complexity and signaling overhead increase
Solution Approach 1:
The patent segments the transform selection into distinct categories (DKT, DGST, DRDST) based on statistical properties, allowing efficient selection without evaluating all possible transforms. The residual blocks are segmented into different types based on their characteristics, and each segment uses a predetermined transform, reducing calculation complexity while maintaining coding efficiency.
Solution Approach 2:
The patent performs preliminary classification of residual blocks based on their statistical properties (isotropic, anisotropic, directional characteristics) before selecting the transform. This preliminary action allows the system to choose from a limited set of transform kernels, reducing the computational burden of transform selection while achieving good adaptation to different block types.
3Adaptability or versatility
If non-separable transform is used to handle complex statistical properties, then adaptability is improved, but calculation complexity and signaling overhead significantly increase
Solution Approach 1:
The patent uses asymmetric graph-based transform kernels (DGST and DRDST) that are specifically designed for anisotropic and directional residual patterns. These asymmetric transforms provide better adaptation to complex statistical properties while maintaining a separable structure, avoiding the high complexity of fully non-separable transforms. The asymmetry in the transform kernel matches the asymmetry in the residual block statistics.
Solution Approach 2:
The patent creates a universal graph-based transform framework that can handle multiple types of residual statistics (isotropic, anisotropic, directional) using a unified mathematical structure. The graph-based transform kernel serves multiple functions by adjusting its parameters and structure based on the input characteristics, providing universal adaptability without requiring separate non-separable transforms for each case, thus reducing signaling overhead.
Data Source
AI summary
A method and a device for decoding a video signal using an adaptive separable graph-based transform. The method includes: receiving a transform index for a target block from the video signal in which the transform index indicates a graph-based transform to be applied to the target block; deriving a graph-based transform kernel corresponding to the transform index; and decoding the target block based on the graph-based transform kernel. The device includes: a parsing unit configured to receive a transform index for a target block from the video signal; and an inverse-transform unit configured to derive a graph-based transform kernel corresponding to the transform index and decode the target block based on the graph-based transform kernel.