Hierarchical Point Cloud Attribute Compression Using Block-Adaptive Transform for Static Situations and Curvature-Grading-Based Refinement for Dynamic Conditions

Recent advances in point cloud capture and immersive applications sparked new interests in point cloud attribute compression. The recognized standard framework lacks full use of point clouds, so we propose a hierarchical point cloud attribute compression method, including a block-adaptive transform and a curvature-grading-based refinement scheme. For static point clouds, the proposed block-adaptive transform first constructs a layered structure. Then, an adaptive transform based on the direction of the surface in the block is introduced. After that, the obtained coefficients are quantized and entropy coding. For dynamic context, the advanced curvature-grading-based refinement classifies voxelized identification points into few high-curvature points and massive low-curvature points, then refines the patches hierarchically based on curvature grades to reduce the compression complexity. Experiments show that our method not only gains clear RD performance benefits than the previous Region Adaptive Haar Transform method in static situations but also achieves faster runtime and improved quality than the video-based point cloud compression solution in dynamic conditions.