| Popis: |
This paper describes a novel method for extracting surface-feature lines, which are curves on surfaces and become C1 boundary curves as fillet edges and boundaries between convex and concave regions as well as sharp edges as C0 boundaries, on meshes. By detecting them, we can proceed high-quality mesh geometry processing such as feature-preserving mesh denoising and simplification, and underlying surface extraction (i.e., segmentation) required for the reverse engineering. Our method is based on the normal tensor framework, and detects both C0 and C1 boundary vertices according to a change of curvature values along the principal directions. Furthermore, our method can be applied to noisy scanned data by performing anisotropic smoothing for the normal tensor. We demonstrate effectiveness of our method by applying it to some CAD models and real-world scanned data. |
