Entropy Coding of Point Cloud Geometry Using Memory Channel

The Point cloud is a popular representation format of 3D objects and scenes. For efficient transmission and storage of point clouds in practice, point cloud compression becomes an attractive research topic for academia and industry. Octree coding is one of the main features for coding the geometry in point clouds, as employed in the latest international standard of Geometry-based Point Cloud Compression (G-PCC). This paper aims to improve the performance of the octree coding in G-PCC with reduced complexity. For this purpose, we employ the neighboring nodes to model contexts for the entropy coding directly. As to neighboring sub-nodes, intermedia states are observed first during the coding process, with a memory channel employed for each state to record the occupancy bits of the already coded sub-nodes with the same state. Then the correlation of the sub-nodes recorded in the same memory channel can be utilized to reduce the spatial redundancy further. Compared to the state-of-the-art GPCC codec, the proposed entropy coding method provides about 1.0% bpp (bit per input point) and 3.5% BD-Rate (Bjøntegaard Delta Rate) reduction under lossless and lossy geometry compression, respectively. Moreover, the proposed method also reduces the complexity.