Abstract
Given a 3D surface mesh M and its corresponding tetrahedral mesh (tet-mesh) T, a novel approach based on graph cut is developed to compute the segmentation of the model T into non-overlapping and compact components with perceptual meaningfulness. The algorithm is simple to implement and completely automatic,and generates a binary tree of components in which the root is the whole 3D tet-mesh T and the leaves are all meaningful sub-tet-meshes. The approach is implemented top-down by partitioning a component into two parts each time until each part is a perceptually meaningful component.
