Fast Denoising of Surface Meshes with Intrinsic Texture

Fast Denoising of Surface Meshes with Intrinsic Texture

Hui Huang1*        Uri Ascher1

  1 University of British Columbia


Figure 1: Comparing with non-iterative bilateral filtering [20]: (a) noisy dragon head model (100 K vertices); (b) smoothed model presented in [20] (80 s on a 1.4 GHz Athlon); (c) smoothed model by our multiscale algorithm (4 iters, K = 1/2, 10 s on our laptop).


We describe a fast, dynamic, multiscale iterative method that is designed to smooth, but not over-smooth, noisy triangle meshes. Our method not only preserves sharp features but also retains visually meaningful fine-scale components or details, referred to as intrinsic texture. An anisotropic Laplacian (AL) operator is first developed. It is then embedded in an iteration that gradually and adaptively increases the importance of data fidelity, yielding a highly efficient multiscale algorithm (MSAL) that is capable of handling both intrinsic texture and mesh-sampling irregularity without any significant cost increase. 


title = {Fast Denoising of Surface Meshes with Intrinsic Texture},
author = {Hui Huang and Uri Ascher},
journal = {Inverse Problems},

volume = {24},

month = {06},

year = {2008},

Downloads (faster for people in China)

Downloads (faster for people in other places)