Video Encoding and Decoding with Adaptive Luma-Chroma Neural Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding and decoding technologies using neural network-based in-loop filters face challenges in flexible processing of different colour components, leading to suboptimal encoding and decoding performance due to uniform filtering approaches and increased computational complexity.
Innovation Solution
A method that determines first and second syntax element identifier information to selectively apply a preset neural network model for filtering based on quantization parameter information, allowing for differential processing of luma and chroma components, thereby optimizing filtering decisions at the block level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a low-complexity preset network model is used for filtering, then computational complexity is reduced, but encoding and decoding performance deteriorates due to uniform processing of different colour components
Solution Approach 1:
The patent segments the filtering process by introducing separate syntax element identifiers for different color components (luma and chroma). This allows selective application of the neural network filter to specific components based on their individual characteristics, rather than uniform processing. The segmentation enables the system to apply filtering only where beneficial, reducing unnecessary computational complexity while maintaining encoding performance.
Solution Approach 2:
The patent implements dynamic filtering control through syntax element identifiers that are determined based on quantization parameter information. The filtering decision is not static but dynamically adjusted per block and per color component, allowing the system to adaptively apply filtering only when it improves encoding performance. This dynamic approach resolves the contradiction by making the filtering process flexible rather than uniformly applied.
2Device complexity
If uniform filtering approach is applied to all colour components, then device complexity is reduced, but encoding and decoding performance deteriorates
Solution Approach 1:
The patent applies local quality by treating different color components (luma and chroma) differently based on their specific characteristics. Separate syntax element identifiers (ph_nnlf_luma_ctrl_flag and ph_nnlf_chroma_ctrl_flag) are introduced to control filtering independently for each component. This allows the filtering process to adapt to the local characteristics of each color component, improving encoding performance without significantly increasing overall system complexity.
Solution Approach 2:
The patent changes the control parameters by introducing component-specific syntax element identifiers that determine filtering application. Instead of a single uniform filtering parameter, the system uses separate parameters for luma and chroma components, allowing independent optimization of filtering for each component type. This parameter differentiation enables better encoding performance while maintaining manageable device complexity.
3Adaptability or versatility
If selective filtering based on syntax element identifiers is implemented, then adaptability for different colour components is improved, but device complexity increases
Solution Approach 1:
The patent achieves universality by designing a filtering control mechanism that can handle multiple color components through a unified framework. The same neural network filter structure is used, but controlled by separate syntax element identifiers that allow the system to adaptively apply filtering to luma, chroma, or both components based on quantization parameter information. This multi-functional approach improves adaptability while avoiding the need for completely separate filtering systems for each component.
Data Source
AI summary
A decoding method is provided and includes that: a bitstream is parsed, and first syntax element identification information of a component to-be-filtered of a current frame or current slice is determined, first syntax element identification information is used for determining whether a component to-be-filtered of each block of the current frame or current slice is filtered by using preset network model; when first syntax element identification information indicates that there are components to-be-filtered of partitioned blocks in the current frame or current slice to be allowed to be filtered by using preset network model, quantization parameter information of the component and second syntax element identification information are determined; and based on second syntax element identification information, quantization parameter information and preset network model, the current block of the current frame or current slice is filtered to obtain a filtered reconstruction value of the component of the current block.