Optimized octree codec for geometry-based point cloud compression

Geometry-based point cloud compression (G-PCC) standard is devoted to generic point clouds and offers excellent performance. However, redundancies are found in the original octree codec of G-PCC. This paper addresses the found issues to optimize the octree codec of G-PCC to be more precise. First, uniform contexts are proposed based on the neighbouring child nodes and nodes for the bitwise mode. Second, we merge the single-child mode into the planar mode. A method of eligibility determination for this new planar mode guided by the required number of coded bits is proposed. In addition, the characteristics of each plane in the point cloud are utilized to design contexts for the new planar mode. Third, we use the exponential moving average to optimize the contexts for the bitwise and planar modes, guaranteeing low memory consumption and high compression performance. Experimental evaluation, performed on a diversity of point clouds in the common test condition of G-PCC, demonstrates that the proposed methods enhance the compression performance of the octree codec in G-PCC with a gain of − 2.5 and − 5.8% for lossless and lossy geometry compression, respectively. At the same time, the computational complexity is reduced. Part of this work has been adopted to the latest G-PCC Edition 2. Compared to methods published in recent years, the improved octree codec of G-PCC also shows the advantages of compressing a diversity of point clouds.