Intra Prediction Model Selection for Boundary-Aware Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in improving intra prediction coding efficiency and accuracy, particularly in handling spatial redundancy and boundary padding, which affect the overall compression and reconstruction quality of video data.
Innovation Solution
Implementing methods such as multi-model extrapolation filter-based intra prediction, intra prediction using multiple hypothesis models, directional padding for picture boundary padding, reference shifting, intra mode derivation with signaled syntax, and histogram of gradients-based directional planar to enhance intra prediction coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional intra prediction methods are used, then device complexity is kept low, but coding efficiency and accuracy deteriorate
Solution Approach 1:
The patent divides the intra prediction process into multiple candidate models (first candidate model and second candidate model), each handling different prediction scenarios. The current block is segmented into different prediction regions based on mode flags, allowing each segment to use the most appropriate prediction method for its characteristics, thereby improving overall coding efficiency while managing complexity through structured division
Solution Approach 2:
The patent dynamically selects between different candidate models based on block characteristics and prediction modes. The decoder determines which candidate model to use by evaluating specific conditions (such as intra prediction mode values and block positions), making the prediction process adaptive rather than static. This dynamic approach allows the system to optimize coding efficiency for different content types without requiring all complexity to be present simultaneously
2Measurement precision
If multiple candidate models are used for intra prediction, then coding accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent changes key parameters such as the intra prediction mode value and block position to determine which candidate model to apply. By using parameter-based selection (mode flags, block coordinates) rather than complex content analysis, the system achieves accurate model selection with minimal computational overhead. The parameter changes drive both the model selection and the prediction process, improving accuracy while controlling complexity
Solution Approach 2:
The patent performs preliminary determination of the candidate model before executing the actual prediction. The decoder evaluates mode flags and block characteristics in advance to select the appropriate model, preparing the prediction path beforehand. This preliminary action avoids complex real-time decisions during prediction execution, reducing computational complexity while maintaining high prediction accuracy
3Manufacturing precision
If boundary padding is applied, then prediction quality at picture boundaries is improved, but processing complexity increases
Solution Approach 1:
The patent applies directional padding specifically to boundary blocks where it is most needed, rather than uniformly to all blocks. The padding direction and method are locally adapted based on the block's position and prediction mode, providing high prediction quality at boundaries while avoiding unnecessary processing in interior regions. This local approach maintains boundary prediction quality while reducing overall processing complexity
Data Source
AI summary
A current block is coded based on an extrapolation filter-based intra prediction according to one of a plurality of candidate models. Each of the plurality of candidate models is associated with a respective filter and includes a plurality of filter coefficients of the respective filter. A reference value is determined based on neighboring samples in a template region of the current block. The one of the plurality of candidate models to predict the current block is determined based on a comparison between the reference value and a threshold value. A clipping range of prediction sample values of the current block is determined. The prediction sample values of the current block are obtained based on the one of the plurality of candidate models. The current block is reconstructed based on a portion of the prediction sample values in the clipping range.