CNN Loop Filtering With Colour Fusion for Video Block-Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding and decoding systems face issues with block-shaped Coding Units (CUs) having different coding parameters, leading to discontinuous boundaries and visual artifacts, affecting the subjective and objective quality of reconstructed pictures and prediction accuracy, and requiring inefficient manual filter design with increased encoding bits.
Innovation Solution
Implementing a Convolutional Neural Network (CNN) filter for in-loop filtering that fuses multiple colour components, reducing calculation complexity and saving coding rate by utilizing the relationship between these components, thus enhancing the quality of reconstructed pictures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional in-loop filters with artificial design are used, then filtering processing can be performed on reconstructed pictures, but the adaptive ability and filtering effect are poor and encoding bits increase
Solution Approach 1:
The system uses the reconstructed picture itself and its color components to generate filtering parameters through automatic analysis, rather than relying on manually designed filters. The filtering parameters are derived from the picture's own characteristics (color component relationships, local statistics), enabling the system to self-adapt to different picture contents without requiring extensive manual configuration or additional encoding bits for filter parameters
Solution Approach 2:
The invention changes the approach from fixed artificial filter parameters to dynamic parameters automatically generated from picture content. By analyzing the relationships between color components and computing filtering parameters based on local picture characteristics, the system adapts filter behavior to match the actual picture content, improving filtering effectiveness without increasing encoding bit expenditure
2Reliability
If CNN filter is used to perform filtering processing, then subjective and objective quality are significantly enhanced, but not all relevant information is fully utilized
Solution Approach 1:
The invention merges multiple color components (Y, U, V) and their interrelationships into the filtering process. By fusing information from different color components and combining it with automatically generated filtering parameters, the system fully utilizes all available relevant information from the reconstructed picture, enabling the CNN filter to achieve optimal performance with complete information utilization rather than partial information
3Productivity
If block-shaped Coding Units with different coding parameters are used, then encoding efficiency is improved, but discontinuous boundaries cause block effects and affect picture quality
Solution Approach 1:
The invention applies different filtering strategies to different regions of the picture based on local characteristics. By analyzing local color component relationships and generating region-specific filtering parameters, the system adapts the filtering strength and method to match local picture content and block boundaries, effectively reducing block effects in regions where they occur while maintaining encoding efficiency benefits from block-shaped CUs
Data Source
AI summary
Provided are a loop filtering implementation method and apparatus, and a computer storage medium. The method includes: obtaining an picture to be filtered, wherein the picture to be filtered is generated from an original picture during the video encoding of a video to be encoded, the video to be encoded includes an original picture frame, and the original picture frame includes the original picture; based on the picture to be filtered, separately obtaining at least two colour components of the picture to be filtered; determining the fusion information of the picture to be filtered, wherein the fusion information is obtained by fusing the at least two colour components; and based on the fusion information, performing loop filtering processing on the picture to be filtered to obtain at least one colour component subsequent to filtering the picture to be filtered.