Point Cloud Denoising using Joint Geometry/Color Graph Wavelets

Abstract

A point cloud is a 3D geometric signal representation associated with other attributes such as color, normal, trans parency. Point clouds usually suffer from noise due to imperfect acquisition systems. Based on the notion that geometry and color are correlated, we present a novel non-iterative framework for point cloud denoising using Spectral Graph Wavelet transform (SGW) that takes advantage of this correlation and performs denoising in the graph frequency domain. The proposed approach is based on the design of a joint geometry and color graph that compacts the energy of smooth graph signals in low-frequency bands. We then apply soft-thresholding to remove the noise from the spectral graph wavelet coefficients. Experimental results show that the proposed technique significantly outperforms state-of-the-art methods.