EfficientMeshDenoising viaRobustNormal Filtering and AlternateVertexUpdating ∗

LI Tao,Jun WANG,Hao LIU,Li-gang LIU

2016-01-01

Abstract:The most challenging problem in mesh denoising is to distinguish features from noise. Based on robust guided normal estimation and alternate vertex updating strategy, we investigate a new, feature-preserving mesh denoising method. To capture local structures around features accurately, we propose a corner-aware neighborhood (CAN) scheme. By combining both overall normal distribution of all faces in a CAN and individual normal influence of the interested face, we give a new consistency measuring method, which greatly improves the reliability of the estimated guided normals. As the noise level lowers, we take as guidance the previous filtered normals, which coincides with the emerging rolling guidance idea. In the vertices updating process, we classify vertices according to filtered normals at each iteration and reposition vertices of distinct types alternately with individual regularization constraints. Experiments on a variety of synthetic and real data indicate that, our method adapts to various noise, both Gaussian and impulsive, no matter whether distributed in the normal direction or randomly, with few triangles