Adaptive Loop Filtering Using Residual-Based Video Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in effectively utilizing adaptive loop filtering to enhance video quality while maintaining efficient compression, particularly in bandwidth and memory resource-limited environments.
Innovation Solution
Implementing adaptive loop filtering techniques that utilize offline-trained fixed filters and online filters, leveraging classifiers such as band-based and residual-based classifiers, to refine spatial neighboring samples and filtering input signals, enhancing video decoding and encoding processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If adaptive loop filtering is applied to enhance video quality, then video quality is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training multiple fixed filters with different parameters before actual video decoding. These pre-trained filters are stored in a lookup table, allowing the system to quickly select and apply the appropriate filter without performing complex real-time learning, thus reducing runtime computational complexity while maintaining high video quality improvement
Solution Approach 2:
The patent utilizes parameter changes by varying filter parameters (such as filter coefficients and operational characteristics) based on classification results. Different classifiers (band-based, residual-based) select appropriate filter parameters from pre-defined sets, enabling adaptive quality enhancement without requiring complex real-time parameter optimization, thereby managing system complexity
2Manufacturing precision
If multiple filters and classifiers are used to improve filtering accuracy, then video quality enhancement is improved, but processing time increases
Solution Approach 1:
The system performs preliminary classification of video blocks using simple criteria (band-based or residual-based) before filter application. This preliminary action quickly identifies the appropriate filter type and parameters in advance, avoiding time-consuming real-time optimization during the actual filtering process, thus reducing overall processing time while maintaining high filtering accuracy
Solution Approach 2:
The patent employs copying by using pre-trained filter models that have been optimized in advance. Instead of training filters during real-time processing, the system copies pre-computed filter parameters from storage based on classification results, significantly accelerating the filtering process while maintaining the accuracy of complex filtering operations
3Productivity
If fixed filters trained offline are used, then computational load during decoding is reduced, but filter adaptability to different video content decreases
Solution Approach 1:
The patent resolves this contradiction through parameter changes by maintaining a diverse set of pre-trained filters with different parameters (coefficients, operational characteristics) corresponding to various video content types. The classification system dynamically selects the appropriate pre-trained filter parameters based on the current video block characteristics, enabling both fast decoding (by avoiding real-time training) and high adaptability (by selecting content-appropriate filters from the pre-defined set)
Data Source
AI summary
Methods and apparatus are provided for video decoding and encoding. In one method, a decoder obtains one or more spatial neighboring samples associated with a current sample, where the one or more spatial neighboring samples are from a residual signal. The decoder then derives an adaptive loop filter (ALF) classifier for an online ALF process, where the ALF classifier utilizes sample values from the residual signal.