Feature-preserving mesh denoising based on guided normal filtering

Abstract

In order to robustly perform tasks based on 3D data model, we propose a feature-preserving mesh denoising algorithm based on the face classification. In the proposed algorithm, the sharp features which play a key role in 3D models are kept unchanged while denoising. The multiscale tensor voting is used to classify the faces into two classes where one is called as feature faces and another as non-feature faces. Feature faces is usually distributed in the neighbourhood of shape edges. Thus these feature faces are key faces in perceptual quality. For processing the faces more efficiently, we propose a search algorithm to find those faces which are close to the feature face and are of similar geometrical properties and then use them to guide the filtering process. The remaining faces are processed by an iteratively joint bilateral filtering. Finally, vertex position is updated according to the filtered face normals. the effectiveness of proposed approach is validated through extensive experiments. Experimental results show the performance is better than the existing methods.