Neural-Network Post-Filtering for Multi-Picture Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies, such as MPEG-2, MPEG-4, HEVC, and VVC, require improvements in coding efficiency.
Innovation Solution
Implementing a neural-network post-filter (NNPF) that processes multiple pictures in a specific order to enhance video coding efficiency and performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional video coding technologies (MPEG-2, MPEG-4, HEVC, VVC) are used, then video compression is achieved, but coding efficiency is insufficient
Solution Approach 1:
The patent replaces traditional mechanical filtering systems with a neural network-based post-filter that uses deep learning models (such as U-Net, ResNet, or Swin Transformer architectures) to process video blocks. This substitution enables more efficient compression by leveraging the neural network's ability to learn optimal filtering parameters from training data, achieving better coding efficiency while maintaining video quality through intelligent artifact reduction and detail preservation.
2Productivity
If neural-network post-filter is applied to multiple pictures, then coding efficiency is improved, but processing complexity increases
Solution Approach 1:
The patent segments the video processing task by applying the neural network post-filter to individual video blocks or slices rather than processing entire pictures or sequences at once. This segmentation approach reduces the computational complexity for each processing step while maintaining the benefits of multi-picture context. The filter can be applied independently to different blocks, enabling parallel processing and reducing overall processing time and resource requirements.
Solution Approach 2:
The patent employs preliminary action by pre-training the neural network model offline using a large dataset of video content, then deploying the trained model for actual video processing. The model weights and architecture are prepared in advance, allowing the processing system to simply execute pre-computed filtering operations during real-time video encoding/decoding. This separates the complex learning task from the execution task, reducing processing complexity during actual video processing while maintaining high coding efficiency.
Data Source
AI summary
Embodiments of the disclosure provide a solution for video processing. A method for video processing is proposed. The method includes: determining, for a conversion between a video unit of a video and a bitstream of the video, a neural-network post-filter (NNPF) is activated for a set of pictures; apply the NNPF to one or more pictures in the set of pictures according to an order; and performing the conversion based on the NNPF.