Chroma Prediction with Dual Models for Nonlinear Luma Correlation

The patent divides the coding block into multiple sub-blocks and applies different prediction models (linear CCLM for some sub-blocks, non-linear for others) based on local luma-chroma correlation characteristics. This segmentation allows the system to maintain simplicity where linear correlation suffices while improving accuracy where non-linear relationships exist, thereby resolving the contradiction between ease of operation and prediction accuracy.

The patent dynamically selects between linear and non-linear prediction models based on local correlation assessment. By evaluating the luma-chroma correlation in different regions and adaptively choosing the appropriate model, the system transitions from a static single-model approach to a dynamic multi-model approach, achieving both operational efficiency and high prediction accuracy across diverse content types.

The patent applies non-linear prediction models only to specific sub-blocks where luma-chroma correlation is determined to be non-linear, rather than applying it universally. This selective segmentation reduces the overall computational burden and device complexity while still achieving improved prediction accuracy in the regions that benefit most from non-linear modeling.

The patent implements local quality by applying different prediction model complexities to different spatial regions based on their specific correlation characteristics. High-complexity non-linear models are applied only where needed (in regions with non-linear correlation), while simpler linear models are used elsewhere, thereby optimizing the balance between prediction accuracy and device complexity through localized model selection.

The patent segments the coding block into sub-blocks and processes each with appropriate model complexity. By identifying and separating regions with non-linear correlation, the system can apply enhanced prediction only where necessary, maintaining high processing speed for the majority of regions while improving overall prediction accuracy through targeted non-linear modeling in critical areas.

The patent applies non-linear prediction modeling partially—only to the extent necessary for regions with non-linear luma-chroma correlation—rather than applying it excessively to the entire block. This partial action approach improves measurement precision where needed while avoiding the full computational overhead that would significantly reduce processing speed, thereby achieving an optimal balance between accuracy and speed.