Saliency-Guided Smoothing For 3d Point Clouds

To efficiently process 3D unorganized point clouds with noise and significant outliers, it is important to smooth the point clouds, but still retain faithfully the original surface geometry as much as possible. In this paper, we present a simple and fast saliency-guided smoothing method of 3D point. The method consists of three stages: firstly, saliency value for each point in noisy point clouds is calculated by site entropy rate method; secondly, robust vertex normal vector is updated by neighboring noisy normal vectors with weight terms related to detected visual saliency metrics; finally, based on least-squares error criterion, vertex position is updated with the integration of vertex normal vector. Analysis and experiments show the advantages of our proposed method over similar methods in the literature on synthetic data.