Intra Prediction Using Distance-Weighted Multi-Block References
Solution Overview
Problem
Current video coding standards, such as ITU-T H.265/HEVC, face limitations in coding efficiency for intra prediction due to reliance on a single row or column of adjacent pixels and a single angular prediction direction, leading to higher bitrates for achieving similar visual quality.
Innovation Solution
An apparatus and method for intra prediction that utilize multiple reference pixels from neighboring blocks, weighted based on distances, to predict pixel values, allowing for a combination of primary and secondary reference pixel values through equations that calculate weights for efficient prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single row or column of adjacent pixels is used for intra prediction, then the device complexity is reduced, but the coding efficiency deteriorates leading to higher bitrates
Solution Approach 1:
The patent divides the reference pixels into multiple groups (first group from top neighboring block, second group from left neighboring block) and processes each group separately with different prediction modes. This segmentation allows the system to capture different spatial directions of pixel correlations without requiring a single complex prediction structure, thereby improving coding efficiency while maintaining manageable device complexity.
Solution Approach 2:
The patent transitions from using a single one-dimensional reference line (either row or column) to utilizing two-dimensional reference regions from multiple neighboring blocks. By incorporating reference pixels from both top and left blocks simultaneously, the prediction process gains an additional spatial dimension, improving the accuracy of pixel value prediction and reducing bitrate requirements.
2Device complexity
If a single angular prediction direction is used, then the computational complexity is reduced, but the prediction accuracy deteriorates
Solution Approach 1:
The patent segments the prediction process into multiple directional components by using separate first and second prediction modes for different groups of reference pixels. This allows each segment to be processed with simpler computational rules while the combination of segments achieves higher overall prediction accuracy, resolving the contradiction between computational complexity and prediction precision.
Solution Approach 2:
The patent merges multiple prediction results from different reference pixel groups and prediction modes to generate the final predicted pixel values. By combining the outputs of first prediction mode (using top reference pixels) and second prediction mode (using left reference pixels), the system achieves superior prediction accuracy without requiring a single complex angular prediction algorithm.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to an apparatus (100) for intra prediction of pixel values of pixels of a current video coding block of a frame of a video signal on the basis of a plurality of reference pixel values of a plurality of reference pixels. The apparatus (100) comprises an intra prediction unit (103) configured to intra predict a pixel value of a current pixel of the current video coding block on the basis of: a first reference pixel value of a first reference pixel of the plurality of reference pixels, wherein the first reference pixel is located in a first neighboring video coding block, a second reference pixel value of a second reference pixel of the plurality of reference pixels, wherein the second reference pixel is located in a second neighboring video coding block, a distance between the current pixel and the first reference pixel, and a distance between the current pixel and the second reference pixel.