Video Block Prediction With Weighted Multi-Reference Fusion

The patent merges multiple reference blocks (first reference block and second reference block) through weighted fusion to generate the prediction block. Instead of selecting only the single best matching block, the method combines information from multiple reference blocks with different weights, where the weights are determined based on matching costs and other factors. This merging approach preserves more reconstructed block information while maintaining manageable complexity through systematic weight assignment.

The patent introduces weight parameters to change the contribution of different reference blocks in the prediction process. By dynamically adjusting these weight parameters based on matching costs and block characteristics, the system optimizes the utilization of reconstructed block information. This parameter-based approach allows flexible control over prediction accuracy without proportionally increasing device complexity.

The patent segments the reference blocks into multiple categories (first reference block and second reference block) with different weights. This segmentation allows the system to process and weight different reference blocks separately before fusion, making the complex weighted fusion process more manageable. By dividing the reference blocks into distinct segments with assigned priorities, the processing complexity is organized and controlled while still achieving improved prediction accuracy.

The patent uses weight parameters to control the contribution of each reference block, providing a systematic way to manage processing complexity. These parameters allow the system to adjust the influence of different reference blocks based on their matching quality, enabling optimized prediction accuracy with controlled computational overhead through parameter-based control rather than exhaustive processing.

The patent performs preliminary weighted fusion of multiple reference blocks before final prediction, rather than directly using the best matching block. This preliminary action of combining multiple reference blocks with appropriate weights prepares a more accurate prediction block that accounts for illumination variations. By performing this fusion operation in advance, the system maintains processing efficiency while improving prediction accuracy for challenging scenarios.

The patent creates a composite prediction block by fusing multiple reference blocks together, similar to creating composite materials with enhanced properties. This composite approach combines the strengths of different reference blocks to produce a prediction that is more robust to illumination changes than any single reference block alone, achieving better accuracy without sacrificing processing speed through efficient fusion algorithms.