Gaussian Curvature Constrained Skeleton Extraction Method Based on MRG

Skeleton representation of 3D models has been widely used for shape similarity comparison,character animation and data compression. This paper presents an optimized skeleton extraction approach by using discrete Gaussian curvature to refine the multi-resolution Reeb graph generation. It first calculates the Gaussian curvature of each vertex which clarifies the convex and concave feature of the local surface. By taking the concave point as constrained points,sub-splitting-lines are generated according to their μ values. A local subdivision method is then adopted to get the sub-regions and the new nodes. The skeleton is finally formed by connecting all the nodes. This optimized skeleton keeps the topological and local features of 3D model very well. A series of experiments demonstrated its efficiency.