3D Mesh Pre-Processing Method Based on Feature Point Classification and Anisotropic Vertex Denoising Considering Scene Structure Characteristics

3D mesh denoising plays an important role in model pre-processing and repair. A fundamental challenge the process is to accurately extract features from noise preserve restore scene structure of model. In this paper, we propose a novel feature-preserving method, which was based on robust guidance normal estimation, accurate feature point extraction anisotropic vertex strategy. The methodology proposed approach as follows: (1) dual weight function that takes into account angle characteristics used estimate normals surface, improved reliability joint bilateral filtering algorithm avoids losing corner structures; (2) filtered facet classify points voting tensor (NVT) raised accuracy integrity classification for noisy model; (3) update strategy triangular denoising: updating non-feature with isotropic neighborhood normals, effectively suppressed sharp edges being smoothed; local geometric constraints, preserved restored while avoided pseudo features. detailed quantitative qualitative analyses conducted synthetic real data show our method can remove various models retain or edge without generating