Multi-Reference-Line Linear Models for Chroma Prediction

The patent divides the chroma prediction process into multiple segments, each corresponding to a different reference line. Instead of using all reference lines simultaneously, the method segments the prediction into multiple candidate models, each derived from a single reference line. This segmentation allows the system to manage computational complexity by processing reference lines individually rather than all at once, while still enabling the selection of the best prediction accuracy through multiple candidates.

The patent implements dynamic selection of reference lines based on prediction performance. The system dynamically determines which reference line provides the best chroma prediction by comparing multiple candidate models and selecting the one with the lowest cost function value. This dynamic approach allows the system to adapt to different content characteristics and maintain high prediction accuracy without permanently committing to the highest computational complexity.

The patent performs preliminary derivation of linear model parameters for multiple reference lines before final selection. By pre-computing candidate models from all available reference lines and evaluating their costs in advance, the system identifies the best prediction model beforehand. This preliminary action ensures that the final prediction uses the most accurate model without requiring additional processing time during the actual encoding/decoding phase, as the selection is already made.

The patent derives linear model parameters for more reference lines than strictly necessary (excessive action), then selects only the best one. This approach allows the system to explore multiple possibilities to ensure optimal prediction accuracy, while the cost function evaluation mechanism ensures that only the most effective model is ultimately applied. The excessive derivation of candidate models is justified by the subsequent selective process that eliminates inferior options.

The patent employs a feedback mechanism through cost function evaluation to select the optimal reference line for chroma prediction. The system derives multiple candidate linear models from different reference lines, evaluates each model's performance using a cost function that measures prediction accuracy, and feeds this evaluation information back to select the best model. This feedback-driven selection process ensures that the most effective prediction model is chosen, improving compression efficiency by reducing the residual energy without permanently increasing model complexity, as only one model is ultimately applied.

The patent changes the parameters of the linear model (slope and offset) based on different reference lines. By deriving separate linear model parameter sets for each reference line and selecting the best set through cost function evaluation, the system adapts the prediction model to match the actual chroma-luma relationship in the current block. This parameter variation allows the system to improve compression efficiency by using the most appropriate model parameters for each specific case without requiring a permanently complex multi-model structure.