Neural Network In-Loop Filtering for Video Coding Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies, such as MPEG-2, MPEG-4, ITU-T.263, ITU-T.264/MPEG-4 AVC, and ITU-T.265 HEVC, require improvements in coding efficiency and complexity reduction for video processing.
Innovation Solution
Implementing a neural network-based in-loop filtering that utilizes various convolution types, kernel size decomposition, multi-scale neural network structures, transformer-based structures, and adaptive parameter settings to enhance filtering performance and reduce complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional video coding technologies are used, then coding efficiency can be maintained at acceptable levels, but filtering performance is insufficient and complexity reduction is limited
Solution Approach 1:
The patent replaces traditional mechanical filtering systems with neural network-based filtering. The neural network filter processes video blocks to reduce blocking artifacts and improve visual quality, substituting conventional linear filtering mechanisms with learned non-linear transformations that achieve superior filtering performance while maintaining coding efficiency.
2Reliability
If complex filtering methods are applied to improve filtering performance, then video quality improves, but computational complexity increases
Solution Approach 1:
The patent employs parameter changes by adaptively adjusting neural network filter parameters based on video content characteristics. Different neural network filters are selected or parameters are modified according to the specific video block properties, allowing the system to maintain high filtering performance while reducing computational complexity by avoiding unnecessary complex operations on all blocks.
Solution Approach 2:
The patent segments the video processing into different blocks and applies selective neural network filtering based on block characteristics. Not all blocks require the same filtering intensity, so the system divides the video into manageable units and applies appropriate filtering operations to each segment, reducing overall computational complexity while maintaining performance.
3Ease of manufacture
If fixed neural network filters are used, then implementation is simple, but adaptability to different video content is limited
Solution Approach 1:
The patent introduces dynamics by making the neural network filter parameters adaptive rather than fixed. The system dynamically adjusts filter parameters or selects different filters based on real-time analysis of video block characteristics, enabling the filtering system to adapt to varying video content while maintaining a relatively simple implementation framework.
Data Source
AI summary
Embodiments of the present disclosure provide a solution for video processing. A method for video processing is proposed. The method comprises: determining, during a conversion between a video unit of a video and a bitstream of the video, a neural network filter according to a rule; applying the neural network filter to the video unit; and performing the conversion based on the filtered video unit.