Semi-supervised 3D Mesh Hierarchical Segmentation

This paper presents a novel and efficient algorithm for the hierarchical mesh segmentation of 3D meshes,which is based on semi-supervised K-means clustering and k-ring strip growing.The new technique consists of three consecutive stages:prominent feature point extraction,pre-segmentation and post-segmentation.The prominent feature points are extracted using multi-dimensional scaling transformation;the original 3D mesh is then segmented initially using the semi-supervised K-means clustering algorithm in the pre-segmentation stage to improve the algorithm's efficiency.According to the pre-segmentation results,we apply the Gaussian curvature and k-ring strip growing algorithms to get a hierarchical segmentation of the mesh.Comparing to the other algorithms,our proposed algorithm can obtain better boundaries,and the regions are harmonic.