Abstract: Decomposition and segmentation of the objects represented by point cloud data become increasingly important for purposes like shape analysis and object recognition. In this paper, we propose a perception based approach to segment point cloud into distinct parts, and the decomposition is made possible of spatially close but geodetically distant parts. Curvature is a critical factor for shape representation, reflecting the convex and concave characteristics of an object, which is obtained by cubic surface fitting in our approach. To determine the number of patches, we calculate and select the critical feature points based on perception information to represent each patch. Taking the critical marker sets as a guide, each marker is spread to a meaningful region by curvature-based decomposition, and also further constraints are provided by the variation of normals. Then a skeleton extraction method is proposed and a label-driven skeleton simplification process is implemented. Further, a semantic graph is constructed according to the decomposed model and the skeletal structure. We illustrate the framework and demonstrate our approach on point cloud data to evaluate its function to decompose object shape based on human perceptions. Meanwhile, the result of decomposition is demonstrated with extracted skeletons. Performance of this algorithm is exhibited with experimental results, which proves its robustness to noise.