Hierarchical Transform Coding for Efficient Point Cloud Attribute Decoding

The patent segments the point cloud data into multiple coding units and further divides each coding unit into multiple nodes organized in a tree structure. This segmentation allows for independent processing of different regions, enabling parallel decoding operations and reducing overall decoding complexity while maintaining compression efficiency.

The patent introduces a hierarchical tree structure dimension to organize the point cloud data, transforming the traditional flat data structure into a multi-level hierarchy. This dimensional change enables more efficient memory access patterns and allows the decoding process to operate at multiple levels of abstraction, reducing computational complexity.

The patent applies inter prediction selectively to different nodes in the hierarchical tree structure based on local characteristics. Rather than uniformly applying prediction to all coefficients, the method adapts the prediction process to local regions, using reference coding units that are spatially and hierarchically relevant, thereby reducing prediction accuracy requirements while maintaining encoding efficiency.

The patent introduces residual values as an intermediary between the predicted coefficients and the actual coefficients. The residual values capture the difference between prediction and actual data, allowing the system to achieve accurate reconstruction without requiring extremely high prediction accuracy, thus balancing encoding efficiency with prediction requirements.