Mesh Segmentation by Local Depth

In this paper, we propose a novel segmentation algorithm based on the local depth, a new measure for capturing the geometrical feature of 3D boundary points. Compared to the curvature feature, the local depth has two advantages: 1) The local depth is in itself defined on a discrete space, while the curvature is defined on a smooth space. Thus, the local depth is more suitable for capturing the geometrical feature of partial boundary on unsmooth 3D surface. 2) The local depth considers a wider range of geometry information than the curvature does, hence it avoids discovering some false boundary, and is able to find more boundaries that the curvature. Based on the local depth, we then introduce the Fast Marching Watersheds algorithm to segment a mesh into representative regions.