Intra Prediction Fusion Using Angular-Planar Weighted Sum
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing High Efficiency Video Coding (HEVC) standard's Intra prediction modes are limited in effectively combining multiple prediction methods to improve coding performance, particularly in handling various block sizes and prediction directions, which can lead to suboptimal prediction accuracy and efficiency.
Innovation Solution
The proposed method combines angular and planar prediction modes using a weighted sum of predictors derived from different prediction methods, incorporating adaptive interpolation processes that utilize multiple reference samples at block corners, and apply different weighting factors based on block dimensions for division-less implementation, thereby generating a final Intra prediction block that enhances prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple Intra prediction modes are combined using weighted sum, then prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent combines multiple Intra prediction modes (planar, angular, and gradient predictions) into a unified prediction framework. The decoder merges these different prediction types by computing a weighted sum of their respective prediction blocks, where the weights are determined by gradient magnitude comparisons. This merging approach allows the system to leverage the strengths of each prediction mode simultaneously, improving overall prediction accuracy while maintaining a systematic integration method.
Solution Approach 2:
The patent dynamically adjusts prediction parameters based on block characteristics. Specifically, it changes the weighting factors applied to different prediction modes according to the gradient magnitudes calculated from reference samples. When vertical gradients are stronger, angular predictions receive higher weights; when horizontal gradients dominate, planar predictions are weighted more heavily. This parameter adaptation allows the system to optimize prediction accuracy for different content types without requiring a fixed complex structure.
2Measurement precision
If adaptive interpolation processes are used with multiple reference samples, then prediction performance is improved, but processing time increases
Solution Approach 1:
The patent implements a selective processing approach where not all prediction modes are always computed. Instead, it first calculates gradient magnitudes from reference samples and uses these to determine which prediction modes are most relevant for the current block. By performing partial actions—computing only the necessary prediction modes based on gradient analysis—the system achieves good prediction performance while avoiding the excessive processing time that would result from always computing all possible prediction modes.
Solution Approach 2:
The patent performs preliminary gradient magnitude calculations on reference samples before committing to specific prediction mode computations. This preliminary action allows the system to pre-determine the weighting strategy and identify which prediction modes will be most beneficial. By preparing this information in advance, the system avoids unnecessary computations during the main prediction phase, thereby reducing overall processing time while maintaining prediction accuracy.
3Productivity
If division-less implementation is applied, then computational efficiency is improved, but precision may be reduced
Solution Approach 1:
The patent replaces the mechanical division operation with bit-shift operations in the weighted sum computation. Instead of performing actual division to calculate weighted averages of prediction blocks, the system uses right-bit-shifts as a substitute for division by powers of two. This substitution maintains computational efficiency by avoiding expensive division operations while preserving sufficient precision through the use of fixed-point arithmetic and appropriate scaling factors in the weighted sum calculation.
Data Source
AI summary
A method and apparatus for video coding using Intra prediction are disclosed. In one method, a first prediction sample in an immediately right column of the current block and a second prediction sample in an immediately below row of the current block are derived using angular prediction. The first prediction sample and a left column reference sample in the same row as the first prediction sample are interpolated to generate a horizontal predictor. The second prediction sample and an above-row reference sample in the same column as the second prediction sample are interpolated to generate a vertical predictor. The vertical predictor and the horizontal predictor are linearly combined to generate an angular-planar prediction sample. In another method, a first predictor is generated using angular prediction and a second predictor is generated using planar prediction. The first predictor and the second predictor are linearly combined to generate a fused Intra predictor.