Point Cloud Attribute Prediction Using Morton-Code Nearest Neighbors

The patent applies preliminary action by pre-calculating and storing Morton codes for all points in the reference frame during the encoding process. These pre-computed Morton codes are stored in a lookup structure that enables O(1) retrieval during decoding, eliminating the need for computationally intensive nearest neighbor searches during the actual prediction process.

The patent replaces the mechanical search process (iteratively comparing distances to find nearest neighbors) with an information-theoretic approach using Morton codes. The Morton code serves as a spatial indexing key that directly maps to neighborhood relationships, substituting complex geometric search operations with simple code lookup and comparison operations.

The patent performs the computationally intensive Morton code calculation and nearest neighbor identification in advance during the encoding process. The results are stored in lookup tables or pre-computed data structures, so that during decoding the process simply involves retrieving pre-computed values, dramatically reducing real-time computational requirements while maintaining optimal prediction accuracy.

The patent segments the point cloud processing into distinct phases: encoding phase where nearest neighbor relationships are established and Morton codes are computed, and decoding phase where these pre-computed relationships are utilized. This temporal segmentation allows heavy computation to be performed when data is available (encoding time) rather than during real-time processing (decoding time).