BVD Magnitude Prediction with Candidate Pruning in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding and decoding technologies face inefficiencies in compressing video data due to limitations in predicting and signaling the magnitude of block vector differences (BVD), leading to increased signaling overhead and reduced compression efficiency.
Innovation Solution
Implementing a method to predict the magnitude symbols of BVD using a BVD predictor, selecting from multiple candidates based on a cost function, and pruning a subset of candidates to improve prediction accuracy and reduce signaling overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple BVD candidates are used to predict magnitude symbols, then prediction accuracy is improved, but signaling overhead increases
Solution Approach 1:
The patent segments the BVD prediction process into multiple stages: first predicting magnitude symbols using a reduced set of candidates (pruned BVD candidates), then separately predicting sign information. This segmentation allows the system to achieve good prediction accuracy for the most important magnitude components while using minimal signaling overhead, as only the residual information needs to be signaled rather than complete BVD values.
Solution Approach 2:
The patent applies different prediction strategies to different components of the BVD. Magnitude symbols are predicted using pruned BVD candidates with lower complexity, while sign information is handled separately with its own prediction mechanism. This local quality approach optimizes the balance between prediction accuracy and signaling overhead for each component according to its specific characteristics.
2Loss of information
If BVD candidates are pruned to reduce complexity, then signaling overhead is reduced, but prediction accuracy may deteriorate
Solution Approach 1:
The patent applies partial action by using a pruned subset of BVD candidates rather than all possible candidates for predicting magnitude symbols. This partial approach reduces the search space and signaling overhead while maintaining sufficient prediction accuracy, as the pruned candidates are selected to be the most representative and informative for the prediction task.
Solution Approach 2:
The patent incorporates feedback mechanisms where the prediction results from pruned BVD candidates are used to refine subsequent predictions and signaling. The system adapts the prediction based on actual decoded values and prediction errors, allowing the pruned candidate set to be optimized iteratively for best performance in reducing overhead while maintaining accuracy.
Data Source
AI summary
Encoding and/or decoding a block of a video frame may be based on a previously decoded reference block in the same frame or a different frame. The reference block may be indicated by a block vector (BV). A block vector difference (BVD) predictor may be used to make predictions about the magnitude of a BVD. The BVD predictor may be selected from BVD candidates, which may be pruned based on costs associated with them. Pruning the BVD candidates may improve additional predictions and thus improve the compression efficiency of the predictions. A motion vector difference (MVD) predictor may be used to make predictions about the magnitude of an MVD. The MVD may be used with a motion vector predictor (MVP) to indicate a motion vector (MV).