3D Mesh Segmentation Based on Multiclass Spectral Graph Partition

Abstract

This paper proposes a novel 3D mesh segmentation algorithm. The idea is to transform the mesh segmentation as a spectral graph partition problem. The nodes of the graph represent faces in the mesh, the weights on the edges connecting nodes represent the similarities between faces. We then exact some geometric features for each face in the mesh to build weighted matrix. By performing multi-way spectral cut of the graph, we can find an optimal segmentation for the input mesh and apply the graph cut method to refine the boundaries between the segments. Finally we evaluate our approach on the Princeton shape benchmark and compare to other existing segmentation algorithm.