Point Cloud Geometry Compression using Parameterized Graph Fourier Transform

Hinata Kirihara, Shoichi Ibuki, Takuya Fujihashi, Toshiaki Koike-Akino, Takashi Watanabe

Abstract: Existing point cloud coding (PCC) methods based on graph signal processing (GSP) have been proposed for dealing with the irregular structures of 3D points. GSP-based PCC can achieve better rate-distortion performance against the typical tree-based PCC, whereas it requires computational resources for searching hyperparameter sets for graph shift operators and performing eigenvalue decomposition for each parameter set. This paper proposes a novel PCC to reduce the time complexity without the degradation of rate-distortion performance. Specifically, we leverage the properties of the parameterized graph shift operator to realize 1) a reduction in the number of hyperparameters, and 2) a decomposition-free hyperparameter search. Evaluations using ShapeNet point cloud dataset show that the proposed scheme achieves almost the same rate-distortion performance with significant reduction on the computational cost compared to the existing graph-based PCCs.