Affine Linear Weighted Intra Prediction Upsampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video encoding techniques, particularly in ALWIP mode, face challenges in accurately predicting video blocks, leading to increased prediction errors at block edges, which require more bits for compression and result in suboptimal rate-distortion tradeoffs.
Innovation Solution
The proposed solution involves filtering boundary reference samples to generate a filtered prediction block by applying an affine model to a subset of edge samples and then filtering these intermediate samples using the full set of edge samples, thereby improving prediction accuracy and reducing prediction errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If affine linear weighted intra prediction is used to improve prediction accuracy, then prediction quality improves, but the complexity of the prediction process increases
Solution Approach 1:
The prediction process is divided into two distinct stages: first generating intermediate samples using an affine model with a subset of boundary reference samples, then filtering these intermediate samples to produce the final prediction block. This segmentation allows the complex affine transformation to be applied only to selected samples rather than all boundary samples, reducing computational complexity while maintaining prediction accuracy.
Solution Approach 2:
Instead of applying the affine model to all boundary reference samples, the method selectively applies it only to a subset of edge samples (fewer than the full set of left and top edge samples). This partial application reduces the computational burden of the affine transformation while still generating sufficient intermediate samples for accurate prediction when combined with the filtering step.
2Measurement precision
If more boundary reference samples are used in the affine model, then prediction accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
The method uses only a subset of boundary reference samples (fewer than the complete set of left and top edge samples) for generating intermediate samples through the affine model. This selective sampling reduces the number of affine transformations needed, thereby decreasing processing time while the subsequent filtering step ensures prediction accuracy is maintained.
Solution Approach 2:
The boundary reference samples are segmented into two groups: a subset used for affine transformation to generate intermediate samples, and the full set used for filtering these intermediate samples. This segmentation allows efficient use of computational resources by applying the complex affine model only where necessary, rather than to all boundary samples.
3Measurement precision
If filtering is applied to all intermediate samples, then prediction accuracy improves, but the number of operations increases
Solution Approach 1:
The filtering operation is applied selectively rather than uniformly to all possible samples. The method filters the intermediate samples generated from the subset of boundary reference samples using the full set of boundary reference samples, achieving accurate prediction without the excessive operations that would result from more comprehensive filtering approaches.
Data Source
AI summary
A device for decoding video data determine that a current block of video data is encoded in an affine linear weighted intra prediction (ALWIP) mode; derives, based on a set of left edge neighboring samples of the current block and a set of top edge neighboring samples of the current block, a subset of left edge samples and a subset of top edge samples; applies an affine model to the subset of left edge samples and the subset of top edge samples to generate an intermediate block of intermediate samples; filters the intermediate samples to generate a final prediction block; decodes the current block of video data based on the final prediction block.