Non-Separable Video Transform for Inter Prediction Luminance Shifts

The patent applies different transform strategies to different frequency components locally. The DC component (top-left coefficient) is preserved without transformation when luminance change is detected, while other frequency components continue to undergo non-separable transform. This local differentiation maintains transform effectiveness for AC components while preserving DC component accuracy in inter prediction blocks with luminance variation.

The patent dynamically changes the transform parameter (whether to transform the DC component) based on the prediction mode. For inter prediction modes, the DC component transformation is disabled, while for intra prediction modes, it remains enabled. This parameter adaptation allows the system to optimize transform effectiveness for each specific prediction scenario.

The patent introduces dynamic adaptability by making the transform behavior dependent on the prediction mode. The system automatically adjusts its transform application based on whether the block uses intra or inter prediction, enabling it to adapt to different luminance characteristics. This dynamic approach allows the same transform mechanism to serve multiple prediction scenarios effectively.

The patent creates a universal transform framework that handles both intra and inter prediction modes through a single adaptive mechanism. By incorporating prediction-mode-based conditional logic, the non-separable transform becomes a multi-functional tool that can effectively process both intra prediction blocks (with high pixel correlation) and inter prediction blocks (with luminance change), eliminating the need for separate transform strategies.