Filtered Intra Block Copy With Multi-Hypothesis Prediction

The patent divides the prediction process into multiple hypotheses, where each hypothesis corresponds to a different filter coefficient set or filter shape. This segmentation allows the system to explore multiple prediction paths simultaneously, selecting the one that maximizes both coding efficiency and prediction accuracy. The merge candidate list is segmented into multiple entries, each representing a different hypothesis, allowing flexible combination of prediction results.

The patent changes the parameters of the FIBC mode by introducing multiple filter coefficient sets and filter shapes as hypotheses. Instead of using a single fixed set of parameters, the system varies these parameters across different hypotheses and uses adaptive weighting factors to combine them. This parameter variation enables the system to achieve higher prediction accuracy while maintaining coding efficiency through the most suitable parameter combination for each block.

The patent performs preliminary action by pre-defining multiple filter coefficient sets and filter shapes as hypotheses before the actual prediction process. These hypotheses are prepared in advance and stored in the merge candidate list, so that during decoding, the system only needs to select and combine pre-prepared hypotheses rather than computing them in real-time. This preliminary preparation reduces the computational burden during actual prediction while maintaining high accuracy.

The patent uses copying by creating multiple copies of the prediction block with different filter coefficients or filter shapes. Each hypothesis generates a copy of the prediction block, and the final prediction is obtained by combining these copies using adaptive weighting factors. This copying approach allows the system to explore multiple prediction possibilities without requiring complex real-time computations, as the copies are based on pre-defined templates.