Neural Network Post-Filter Signaling in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as ITU-T H.264, H.265, and H.266, lack efficient methods for signaling neural network post-filter parameter information, which limits the effectiveness of post-processing techniques for improving video quality.
Innovation Solution
The proposed techniques involve signaling neural network post-filter parameter information through specific syntax elements, such as the number of output pictures for interlaced input pictures and picture rotation, within the video coding bitstream, enabling better integration of neural network-based post-filters for enhanced video quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional video coding standards are used without neural network post-filter signaling, then device complexity is reduced, but video quality and compression efficiency deteriorate
Solution Approach 1:
The patent segments the video coding process by introducing a separate neural network post-filter stage that operates independently after the traditional coding pipeline. This allows the neural network to be selectively applied only when beneficial, maintaining simplicity for standard cases while enabling quality enhancement when needed.
Solution Approach 2:
The patent introduces syntax elements as intermediaries that carry neural network post-filter parameters through the bitstream. These syntax elements act as a bridge between the traditional video coding standards and the neural network processing, enabling integration without requiring fundamental changes to either system.
2Loss of information
If neural network post-filter parameters are not signaled in the bitstream, then device complexity is reduced, but loss of information occurs for post-processing techniques
Solution Approach 1:
The patent extracts only the essential neural network post-filter parameters that are needed for reconstruction, rather than transmitting complete model information. This selective extraction of critical parameters (such as deinterlacing mode, rotation flags, and scaling factors) reduces information loss while keeping the bitstream overhead minimal.
Solution Approach 2:
The patent changes the representation of neural network parameters from detailed model specifications to compact syntax elements that indicate key processing characteristics. This parameter transformation enables efficient transmission of essential information without the complexity of full model description.
3Productivity
If neural network post-filters are integrated into video coding standards, then compression efficiency is improved, but ease of operation deteriorates due to implementation complexity
Solution Approach 1:
The patent designs the syntax element structure to be universal and compatible with existing video coding standards. The same syntax framework can accommodate different neural network models and processing types (deinterlacing, super-resolution, denoising), making the system multi-functional while maintaining ease of implementation across various scenarios.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A device may be configured to perform filtering based on information included in a neural network post-filter characteristics message. In one example, the neural network post-filter characteristics message includes a syntax element indicating a number of output pictures generated for each interlaced input picture. In one example, the neural network post-filter characteristics message includes a syntax element specifying whether an input picture is rotated.