Adaptive Intra-Inter Frame Prediction Filtering via RDO
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding methods, such as those in the HEVC standard, have limited filtering techniques that only target a few intra prediction modes, do not fully utilize spatial correlation between pixels, and do not consider inter prediction, resulting in suboptimal prediction accuracy and encoding efficiency.
Innovation Solution
A filtering method for intra-frame and inter-frame prediction that adaptively selects filtering through Rate-distortion Optimization (RDO) using a generalized Gaussian distribution function-based filter, applicable to all prediction modes, which determines whether to filter prediction blocks and adjusts filter coefficients based on prediction block size to improve accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing filtering techniques are used for intra prediction, then filtering can be applied to specific prediction modes, but the filtering range is very limited and performance is constrained
Solution Approach 1:
The patent extends the filtering technique to be universally applicable across all intra prediction modes and inter prediction modes, not just limited to specific modes. The filter can be applied to prediction blocks of various sizes (e.g., 4x4, 8x8, 16x16, 32x32) and different prediction types, making the filtering solution multi-functional and broadly applicable throughout the video coding process
Solution Approach 2:
The patent introduces adaptive filtering where the filter parameters and application decisions are dynamically determined based on rate-distortion optimization (RDO). The filtering is not statically applied to all blocks but adaptively selected for each prediction block based on whether it improves the rate-distortion performance, making the filtering process dynamic and context-dependent
2Device complexity
If simple filtering is applied, then computational complexity is reduced, but spatial correlation between pixels is not fully utilized
Solution Approach 1:
The patent employs a generalized Gaussian distribution function to model pixel intensity variations, using parameters (shape parameter, scale parameter) that can be adapted to different prediction blocks. This parameter-based approach allows the filter to capture various spatial correlation patterns without requiring overly complex processing, balancing computational feasibility with accurate modeling of pixel relationships
3Measurement precision
If intra prediction filtering is implemented, then intra prediction accuracy improves, but inter prediction cases are not considered
Solution Approach 1:
The patent creates a universal filtering solution that works for both intra prediction and inter prediction. The same generalized Gaussian distribution-based filter and RDO decision framework are applied regardless of whether the prediction is intra or inter type, making the solution versatile across all prediction scenarios in video coding
4Ease of manufacture
If fixed filter coefficients are used, then implementation is simpler, but prediction blocks of various sizes cannot be optimally filtered
Solution Approach 1:
The patent uses the generalized Gaussian distribution function with adjustable parameters that can be tailored to different prediction block sizes. Different block sizes (4x4, 8x8, 16x16, 32x32) can have optimized parameter sets derived from training data, allowing each size to be filtered with parameters specifically suited to its characteristics while maintaining the same fundamental filtering framework
Solution Approach 2:
The patent performs preliminary training to determine optimal filter parameters for different prediction block sizes before actual video coding. Training data is processed in advance to derive the generalized Gaussian parameters that best fit each block size category, so that during encoding/decoding, pre-determined parameters can be applied without real-time computation, balancing optimization with implementation efficiency
Data Source
AI summary
The present application provides a filtering method for intra-frame and inter-frame prediction. For a pixel value obtained by prediction, whether to filter a prediction block is self-adaptively determined by means of an RDO decision; and a filter is established based on a generalized Gaussian distribution function, and different filters are used for prediction blocks of different sizes, and filter parameters are obtained by training the prediction blocks of each size. The method comprises performing RDO on each coding unit on a coding end, and firstly performing intra-frame prediction or inter-frame prediction on each coding unit; then filtering the prediction block using the filtering method of the present invention, and determining whether to use prediction block filtering by means of an RDO decision; on a decoding end, configuring different filtering coefficients for different prediction block sizes; and decoding according to a flag. Therefore, the problem of prediction block distortion is solved, and the prediction precision can be improved, improving the coding efficiency of an intra-frame and inter-frame prediction block.