Feature-Preserving Anisotropic Smoothing for Meshes with Large-Scale Noise

This paper presents a smoothing method which preserves features for a triangular mesh even when large-scale noise are included because of measurement errors. First, scale-dependent discrete Laplacian is introduced along with boundary Laplacian to deal with an open mesh. Then, a method for feature detection which uses the values by these Laplacians is constructed. Furthermore, anisotropic diffusion is proposed which determines suitable parameters from the values for preserving features. Finally a method is presented which discriminates features from large-scale noise by generating graph of feature lines. Effectiveness of the methods is shown by the experiment results of well-smoothed meshes with their features preserved.