Video Processing With CCCM Gradient Prediction for Coding Efficiency

The patent applies convolutional cross-component modeling to transform the relationship between luma and chroma components through gradient-based parameter estimation. By modeling chroma gradients as functions of luma gradients with learned parameters (slope and intercept), the system achieves more accurate chroma prediction, improving coding efficiency while maintaining video quality.

The patent introduces luma gradients as an intermediary to predict chroma gradients. Instead of directly compressing chroma data, the system uses luma gradient information as a mediator to derive chroma predictions, leveraging the strong correlation between luma and chroma components to achieve better compression efficiency.

The patent segments the chroma prediction process into distinct gradient calculations for different directions (horizontal, vertical, and diagonal). By dividing the prediction task into directional components, the system can apply selective modeling only where needed, reducing overall computational complexity while maintaining coding efficiency.

The patent applies convolutional cross-component modeling selectively to specific video blocks based on gradient characteristics. Instead of applying the complex model uniformly across all blocks, the system identifies blocks where gradient-based prediction provides significant improvement and applies the model only to those cases, reducing overall computational burden.