Hybrid Octree-Plane Point Cloud Geometry Coding

Point clouds provide an efficient way of representing 3D data for applications such as virtual or augmented reality, free-viewpoint video, and gaming. However, this rich, realistic and immersive representation demands a very high amount of data and thus efficient point cloud coding solutions are increasingly needed. A promising way to represent point clouds is through octree partitioning, which provides good compression performance with low complexity, as well as level-of-detail scalability. However, other data representations may offer additional benefits to the octree data structure. The goal of this paper is to propose a point cloud geometry coding solution that leverages the adaptive octree partitioning, and enhances it with a novel coding mode. This mode exploits a plane representation for leaf nodes at different layers of the octree. This corresponds to a hybrid solution where two point cloud representation models are combined. The proposed approach can outperform octree based solutions that are currently considered for standardization and it also provides significant compression gains against a static octree solution (average BD-rate gains of 35% are reported).