Adaptive Loop Filter Shape Switching With Extended Taps for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently utilizing adaptive loop filters (ALFs) to enhance video quality, particularly in handling different color components and formats, leading to suboptimal bandwidth usage and video quality.
Innovation Solution
The implementation of extended taps in ALFs that utilize spatial neighbor samples from reconstruction stages of deblocking filters, sample adaptive offset filters, and bilateral filters, along with conditional filter shape switching based on configuration settings and slice types, to improve filtering efficiency and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional ALF with fixed filter shapes is used, then device complexity is reduced, but video quality and filtering adaptability deteriorate
Solution Approach 1:
The patent implements dynamic filter shape selection where the ALF can switch between different filter shapes (e.g., 5×5 diamond, 7×7 diamond, 5×5 square) based on picture type (I-slice, P-slice, B-slice) and color component characteristics. This dynamic adaptation allows the filtering process to optimize video quality for different content types without requiring manual configuration, resolving the contradiction between video quality and device complexity.
Solution Approach 2:
The patent applies different filter shapes and sizes to different color components (luma, Cb, Cr) and different picture regions based on their specific characteristics. For example, extended 7×7 filters may be applied to luma components in I-slices while using standard 5×5 filters in B-slices. This localized optimization improves overall video quality without uniformly increasing complexity across all processing paths.
2Measurement precision
If extended taps are added to ALF, then filtering precision is improved, but processing time increases
Solution Approach 1:
The patent implements extended taps (e.g., 7×7 or 9×9 filter kernels) only when beneficial, specifically for luma components in I-slices and P-slices where higher precision improves video quality. For B-slices and chroma components, standard 5×5 filters are used. This partial application of extended action achieves filtering precision improvement without uniformly increasing processing time across all video data.
Solution Approach 2:
The filtering process is segmented by picture type and color component, with extended taps applied only to specific segments (luma in I/P-slices) rather than uniformly to all segments. This segmentation allows the system to achieve high filtering precision where needed while maintaining efficient processing for other segments, resolving the time-quality tradeoff.
3Adaptability or versatility
If conditional filter shape switching is implemented, then adaptability is improved, but device complexity increases
Solution Approach 1:
The patent implements dynamic filter shape switching controlled by picture type indicators (I-slice, P-slice, B-slice) and color component identifiers. The control logic automatically selects appropriate filter shapes based on these parameters without requiring complex decision trees or manual configuration. This dynamic control based on simple parameters achieves high adaptability while keeping the control logic relatively simple and implementable.
Solution Approach 2:
The patent changes filter parameters (shape, size, tap count) based on picture type and color component parameters. For example, when picture type is I-slice and component is luma, the filter shape parameter is set to 7×7 diamond; when picture type is B-slice, it switches to 5×5 square. This parameter-based control achieves adaptability through simple parameter changes rather than complex structural modifications.
Data Source
AI summary
A mechanism for processing video data is disclosed. The mechanism includes determining to use at least one extended tap in an adaptive loop filter (ALF). A conversion can then be performed between a visual media data and a bitstream based on the ALF. The ALF may also employ a conditional filter shape switch.