Adaptive Reference Filtering for Video Coding Sharpness Mismatches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video encoding and decoding methods, particularly in motion compensation and multi-view video coding, face inefficiencies due to discrepancies in sharpness/blurriness between frames and mismatched camera settings, leading to suboptimal prediction signals.
Innovation Solution
Adaptive reference filtering is employed to calculate and apply filters to reference pictures, compensating for frame mismatches by generating filtered reference pictures for improved prediction, using techniques like Gaussian Mixture Models for filter determination and differential coding for efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If motion compensation is used to exploit temporal redundancy, then coding efficiency is improved, but prediction accuracy deteriorates when there is sharpness/blurriness discrepancy between current and reference frames
Solution Approach 1:
The patent applies dynamic adaptive filtering where filter parameters are adjusted based on the detected sharpness/blurriness characteristics of reference frames. Instead of using a fixed filtering approach, the system dynamically selects or modifies filter parameters to match the specific conditions of each reference frame, thereby maintaining prediction accuracy while preserving coding efficiency benefits.
Solution Approach 2:
The patent changes filtering parameters adaptively based on the sharpness/blurriness properties of reference frames. By detecting the focus characteristics and adjusting filter parameters accordingly, the system compensates for sharpness mismatches between reference and current frames, resolving the contradiction between maintaining coding efficiency and ensuring prediction accuracy.
2Productivity
If disparity compensation is applied in multi-view video coding, then coding efficiency is improved, but prediction quality deteriorates due to illumination, color, and focus mismatch among different views
Solution Approach 1:
The patent applies local adaptive filtering where different filter parameters are used for different regions of the image based on local sharpness characteristics. This allows the system to compensate for focus mismatches in specific regions without affecting the entire frame, thereby maintaining prediction quality while preserving the coding efficiency gains from disparity compensation.
Solution Approach 2:
The patent adaptively adjusts filter parameters based on the detected focus and sharpness characteristics of different views. By changing filtering parameters to match the specific conditions of each view, the system compensates for illumination, color, and focus mismatches, resolving the contradiction between coding efficiency and prediction quality in multi-view video coding.
3Device complexity
If fixed interpolation filters are used for motion compensation, then device complexity is reduced, but prediction accuracy deteriorates when focus changes occur between frames
Solution Approach 1:
The patent transitions from fixed interpolation filters to dynamic adaptive filters that adjust their parameters based on detected focus changes. The system monitors sharpness characteristics and modifies filter behavior accordingly, maintaining prediction accuracy during focus changes while keeping the overall system complexity manageable through efficient adaptation mechanisms.
Solution Approach 2:
The patent changes interpolation filter parameters adaptively based on focus detection results. By adjusting filter parameters to match the current focus state, the system compensates for focus changes between frames, resolving the contradiction between using simple fixed filters and maintaining high prediction accuracy when focus varies.
Data Source
AI summary
There are provided methods and apparatus for adaptive reference filtering. An apparatus includes an encoder for encoding at least one picture. The encoder performs adaptive filtering of at least one reference picture to respectively obtain at least one filtered reference picture, and predictively codes the at least one picture using the at least one filtered reference picture. The at least one reference picture is a picture wherein at least one sample thereof is used for inter-prediction either of subsequent to the at least one sample being applied to an in-loop filter or in an absence of the at least one sample being applied to an in-loop filter.