Neural Post-Processing Filter for Multi-Purpose Video Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video processing technologies, such as MPEG-2, MPEG-4, ITU-T.263, ITU-T.264/MPEG-4 AVC, ITU-T.265 HEVC, and VVC, require enhancements in functionality for improved video coding and decoding capabilities.
Innovation Solution
Implementing a neural-network post-processing filter (NNPF) that allows for conversions between videos and bitstreams, enabling the generation of pictures with different types of information and supporting additional inputs, thereby extending and enhancing the functionality of video processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional video coding techniques (MPEG-2, MPEG-4, H.264, H.265, VVC) are used, then video compression and decoding functionality is maintained, but the functionality cannot be further improved or extended
Solution Approach 1:
The patent applies universality by designing the NNPF to perform multiple different picture generation purposes using a single unified neural network filter structure. The filter can generate various types of pictures (e.g., enhanced quality pictures, interpolated pictures, super-resolved pictures) based on different configuration parameters, eliminating the need for separate processing units for each function and thereby extending functionality without proportionally increasing system complexity.
2Adaptability or versatility
If a neural-network post-processing filter is applied to generate pictures with different types of information, then the functionality of NNPF is extended and enhanced, but the complexity of the processing system increases
Solution Approach 1:
The patent applies dynamics by making the NNPF configurable and adaptable through parameters indicated in the bitstream. The filter can dynamically adjust its processing mode, input picture selection, and output picture type based on these parameters, allowing a single flexible structure to handle multiple functions rather than requiring multiple fixed-function filters, thus extending functionality while controlling complexity.
Solution Approach 2:
The patent utilizes parameter changes by encoding configuration parameters in the bitstream that control the NNPF's behavior. By changing parameters such as the type of picture to generate, the reference picture selection, and the processing mode, the same neural network filter can perform different functions. This parameter-driven approach allows functionality extension without adding hardware complexity, as the same physical filter adapts its behavior through software-controlled parameters.
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: performing a conversion between a video and a bitstream of the video, wherein a neural-network post-processing filter (NNPF) is applied on a first picture associated with the video, the bitstream comprises a first indication indicating a purpose of the NNPF, a first candidate of a plurality of candidates for the purpose is generating a second picture corresponding to the first picture, and a type of information comprised in the second picture is different from the first picture.