Intra Prediction With Multiple Reference Lines and Lower Signaling Overhead

The patent segments the reference lines into multiple categories (first reference lines and second reference lines) with different roles. First reference lines are used for initial prediction while second reference lines are derived from prediction errors. This segmentation allows the system to manage complexity by processing reference lines in stages rather than treating them uniformly.

The patent performs preliminary derivation of second reference lines based on prediction errors before using them for final prediction. By pre-processing and deriving additional reference lines from the prediction errors of first reference lines, the system prepares multiple reference candidates in advance, improving prediction performance without increasing real-time computational complexity.

The patent extracts and derives second reference line indices from prediction errors rather than transmitting them directly. By taking out the essential information (prediction errors) and transforming it into derived reference line indices, the system reduces the amount of raw data that needs to be signaled while maintaining encoding efficiency.

The patent changes the representation parameters of reference line indices by deriving them from prediction errors. Instead of directly signaling multiple reference line indices, the system signals prediction errors and derives the indices from these errors, changing the parameter space from direct index signaling to error-based derivation, which reduces overhead.