Abstract: 3D shape segmentation is an important problem in 3D shape analysis.In order to get consistency results under various shape changes of a non-rigid object,this paper presents a non-rigid 3D shape segmentation approach based on diffusion geometry.First,the vertices with local extremes of wave kernel signature(WKS) are taken as critical points.Then,taking these critical points as the initial cluster centers,K-means clustering approach is applied to get the segmentation results.The extensive experiments show that the proposed approach can get consistent segmentation results for a non-rigid 3D object under various postures,and is robust to noises and holes as well.