Weighted Intra Prediction for Non-Rectangular Video Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in accurately predicting and coding non-rectangular blocks due to inconsistent motion between foreground and background entities, leading to poor prediction and coding performance.
Innovation Solution
A method is provided to configure reasonable weight values for each pixel position in a current block, using reference weight values from surrounding positions to improve prediction accuracy and performance by approximating prediction values to raw pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rectangular partition is used for motion prediction, then the coding process is simple, but the prediction accuracy deteriorates when motions of foreground and background entities are inconsistent
Solution Approach 1:
The current block is divided into multiple non-rectangular sub-blocks based on the contour of objects, allowing different motion predictions to be applied to different semantic regions. This segmentation enables accurate handling of objects with inconsistent motion (foreground vs background) while maintaining coding efficiency through contour-based partitioning.
Solution Approach 2:
Different prediction modes and weight values are applied to different sub-blocks based on their local characteristics. Each sub-block corresponding to a specific object or region receives customized motion prediction parameters, improving local prediction accuracy while maintaining overall coding performance.
2Adaptability or versatility
If non-rectangular sub-blocks are used for prediction, then the motion representation improves, but the prediction performance and coding performance deteriorate
Solution Approach 1:
The patent introduces weight values as an additional dimension to the prediction process. Instead of relying solely on geometric sub-block shapes, weight values are applied to pixel-level predictions to refine the output. This dimensional enhancement allows standard rectangular partitions to achieve non-rectangular prediction effects, improving reliability while maintaining adaptability.
Solution Approach 2:
The prediction process combines multiple components: rectangular partitioning structure, contour-based sub-block division, and pixel-level weight values. This composite approach integrates the simplicity of rectangular grids with the precision of contour-based segmentation and the flexibility of weighted pixel adjustments, achieving both good prediction performance and coding efficiency.
3Measurement precision
If weight values are configured for each pixel position, then the prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
Weight values are configured locally for each pixel position based on its specific characteristics and the local motion patterns. This localized approach allows high prediction accuracy where needed while avoiding unnecessary computations in uniform regions, balancing precision and complexity.
Solution Approach 2:
The patent dynamically adjusts weight values as parameters based on motion characteristics, object contours, and prediction modes. By changing these parameters adaptively rather than using fixed values, the system achieves high prediction accuracy while maintaining computational efficiency through parameter optimization.
Data Source
Figure 1
Figure 2A~2C
Figure 3~4A
AI summary
Encoding and decoding methods and apparatuses, coder-side devices and decoder-side devices are provided, the method includes: obtaining a weighted prediction angle and weight configuration parameters of the current block; where the weight configuration parameters include a weight transform rate and a weight transform start position; configuring reference weight values for surrounding positions outside the current block according to the weight configuration parameters; determining a surrounding matching position to which the pixel position points from the surrounding positions outside the current block based on the weighted prediction angle; determining a target weight value of the pixel position based on the reference weight value associated with the surrounding matching position, determining an association weight value of the pixel position based on the target weight value of the pixel position; determining a first prediction value of the pixel position based on a first prediction mode, determining a second prediction value of the pixel position based on a second prediction mode; determining a weighted prediction value of the pixel position based on the first prediction value, the target weight value, the second prediction value and the association weight value.