Edge-Adaptive Graph Transform for Video Coding Efficiency

The patent applies dynamics by making the transform basis functions adaptive rather than fixed. The GBT (Graph-Based Transform) dynamically adjusts its basis functions according to the detected edge direction in the residual signal. The transform rotates its basis vectors to align with the dominant edge orientation, allowing the same transform framework to efficiently handle vertical, horizontal, diagonal, and other edge directions without requiring multiple fixed transforms.

The patent changes the parameters of the transform by rotating the basis functions according to the detected edge angle. The transform parameters (rotation angle, basis function orientation) are adjusted based on the edge direction detected in the residual signal. This parameter adaptation allows the transform to optimize its performance for each specific edge configuration rather than using a fixed parameter set.

The patent applies segmentation by dividing the transform process into multiple stages: first detecting edges in the residual signal, then applying GBT only to regions with significant edge content. This selective application avoids processing entire large transform units uniformly, reducing the effective computational burden while maintaining coding efficiency for high-resolution video where large transform units are necessary.

The patent applies local quality by adapting the transform characteristics to local edge properties rather than applying a uniform transform across the entire transform unit. The GBT rotates its basis functions to match the local dominant edge direction, providing locally optimized transformation that improves coding efficiency without requiring excessive processing resources for the entire high-resolution block.