Neural Network Post-Filter Signaling With Quantization-Based Strength Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards lack efficient methods for signaling neural network post-filter parameter information, which can enhance video quality but are not adequately integrated into current video coding frameworks.
Innovation Solution
Incorporating techniques for signaling neural network post-filter parameter information through syntax elements in the video coding process, allowing for the update and control of neural network post-filters, including filtering strength control based on quantization parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If neural network post-filter parameter information is integrated into video coding standards, then video quality is enhanced, but device complexity increases
Solution Approach 1:
The neural network post-filter parameters are nested within existing video coding bitstream structures. The filter parameters are embedded as syntax elements within the video data structure, allowing the neural network filter to be integrated without creating a separate complex system. This nesting approach enables quality enhancement while minimizing the increase in overall system complexity by leveraging existing coding frameworks.
Solution Approach 2:
The patent modifies existing video coding parameters by introducing neural network post-filter strength control information as additional syntax elements. These parameter changes allow the system to adjust filter strength based on quantization parameters and other video characteristics, enhancing video quality while maintaining compatibility with existing coding standards through controlled parameter extensions rather than fundamental system redesign.
2Manufacturing precision
If neural network post-filter control information is added to video bitstream, then image processing is improved, but data requirements increase
Solution Approach 1:
The neural network post-filter parameters are applied locally to specific regions or blocks of video data rather than uniformly across the entire video stream. The filter strength control information is selectively encoded based on local video characteristics and quantization parameters, allowing image processing improvements in critical regions while minimizing the overall data overhead by avoiding unnecessary filter application throughout the entire bitstream.
Solution Approach 2:
The patent implements partial application of neural network post-filtering by controlling filter strength based on quantization parameters. Rather than applying the full neural network filter to all video data, the system applies filtering selectively where needed, using partial action to balance image processing quality improvement with data requirement control. This partial application approach prevents excessive data overhead while maintaining effectiveness in critical areas.
3Adaptability or versatility
If filtering strength is controlled by quantization parameters, then compatibility with video coding standards is maintained, but filter performance may be limited
Solution Approach 1:
The patent implements dynamic filter strength control where the neural network post-filter parameters are adjusted based on quantization parameters and other video characteristics. This dynamic adaptation allows the system to maintain compatibility with video coding standards while optimizing filter performance for different video conditions. The filter strength is not fixed but dynamically adjusted to balance standard compliance with performance requirements across varying video quality levels and coding conditions.
Data Source
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 auxiliary input data is present in an input tensor of the neural network post-filter. The device may derive a filtering strength control value array where each entry in the array is based on a quantization parameter of a respective input picture for the neural network post-filter.