Intra Prediction Reference Handling for Accurate Video Block Decoding

The patent segments the intra prediction mode candidate list into two distinct groups: MPM (Most Probable Mode) candidates and non-MPM candidates. This segmentation allows the decoder to first attempt to derive the intra prediction mode from the smaller MPM candidate group, and only if necessary, expand to the full candidate list. This hierarchical segmentation improves accuracy by providing more specific candidates while managing complexity through conditional processing.

The patent implements dynamic block division where the current block is adaptively divided into sub-blocks based on specific conditions (block size, prediction mode type). This dynamic approach allows the system to process only necessary blocks with fine-grained division, while leaving other blocks undivided. This conditional dynamic processing improves encoding efficiency for complex regions while avoiding unnecessary processing time for simpler regions.

The patent uses available neighboring pixels as intermediaries to replace unavailable reference pixels in the prediction process. When reference pixels are unavailable (e.g., at block boundaries or in unavailable regions), the system derives alternative reference values from adjacent available pixels through interpolation or copying. This intermediary approach maintains prediction accuracy by providing substitute reference data while minimizing information loss through intelligent selection of replacement sources.