Local Surface Descriptor for Geometry and Feature Preserved Mesh Denoising

3D meshes are widely employed to represent geometry structure of shapes. Due limitation scanning sensor precision and other issues, inevitably affected by noise, which hampers the subsequent applications. Convolultional neural networks (CNNs) achieve great success in image processing tasks, including 2D denoising, have been proven own capacity modeling complex features at different scales, is also particularly useful for mesh denoising. However, due nature irregular structure, CNNs-based denosing strategies cannot be trivially applied meshes. To circumvent this limitation, paper, we propose local surface descriptor (LSD), able transform deformable around a face into grid representation thus facilitates deployment CNNs generate denoised normals. verify superiority LSD, directly feed LSD classical Resnet without any complicated network design. The extensive experimental results show that, compared state-of-the-arts, our method achieves encouraging performance with respect both objective subjective evaluations.