Bilateral Filter Weights Based on Prediction Mode
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding techniques face challenges in improving the quality of decoded video due to over-smoothing caused by deblocking filtering and limitations in deriving optimal weights for bilateral filtering, particularly in inter-predicted blocks and across transform unit boundaries.
Innovation Solution
The proposed solution involves deriving bilateral filter weights based on mode information, such as prediction modes, and applying these weights to neighboring samples, including those outside transform unit boundaries, to enhance the filtering process, thereby improving video coding quality and achieving better rate-distortion tradeoffs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deblocking filtering is applied to reduce blocking artifacts, then video quality is improved, but over-smoothing occurs particularly in inter-predicted blocks
Solution Approach 1:
The patent applies different filtering strategies based on the prediction mode of each block. Intra-predicted blocks receive standard bilateral filtering, while inter-predicted blocks receive reduced or no filtering. This local differentiation prevents over-smoothing in inter-predicted blocks while maintaining quality improvement in intra-predicted blocks.
Solution Approach 2:
The patent modifies the bilateral filter parameters (weights, sigma values) based on the prediction mode and transform unit boundaries. By changing these parameters dynamically according to block characteristics, the system avoids over-smoothing in specific regions while maintaining filtering benefits where appropriate.
2Ease of manufacture
If bilateral filter weights are derived without considering prediction modes, then the filtering process is simple, but optimal weights cannot be achieved for different block types
Solution Approach 1:
The patent makes the filter weights dynamic by deriving them based on prediction modes and transform unit boundaries. The weight calculation adapts to different block types (intra vs. inter predicted) and spatial locations, transforming a static filtering process into a dynamic one that optimizes performance for each specific case.
Solution Approach 2:
Different weights are assigned to different regions based on their prediction mode and boundary characteristics. The system recognizes that intra-predicted blocks, inter-predicted blocks, and boundary regions have different filtering requirements, and applies locally optimized weights accordingly.
3Device complexity
If transform unit boundaries are respected in filtering, then processing is straightforward, but filtering performance deteriorates across boundary regions
Solution Approach 1:
The patent segments the filtering process into different operations based on transform unit boundaries. Samples inside transform units receive one type of filtering, while samples at boundaries receive different filtering treatment. This segmentation allows the system to handle boundary regions specially to maintain filtering performance across boundaries.
Solution Approach 2:
The patent applies different filtering characteristics to boundary regions compared to interior regions. By recognizing the special nature of transform unit boundaries and applying locally optimized filtering parameters to these regions, the system maintains high filtering performance across boundaries without excessive complexity.
Data Source
AI summary
A device for decoding video data determines mode information for a current block of a current picture of the video data; derives weights for use in a bilateral filter based on the mode information for the current block; applies the bilateral filter to a current sample of the current block by assign the weights to neighboring samples of the current sample of the current block and the current sample of the current block and modifying a sample value for the current sample based on sample values of the neighboring samples, the weights assigned to the neighboring samples, the sample value for the current sample, and the weight assigned to the current sample; and based on the modified sample value for the current sample, outputs a decoded version of the current picture.