國立臺灣大學資訊工程學系Wu, Fu-CheFu-CheWuMa, Wan-ChunWan-ChunMaLiou, Ping-ChouPing-ChouLiouLiang, Rung-HueiRung-HueiLiangOuhyoung, MingMingOuhyoung2006-09-272018-07-052006-09-272018-07-052003http://ntur.lib.ntu.edu.tw//handle/246246/20060927122916679351Skeleton should be the reduced object representation that conforms to human visual perception, and is very useful in applications such as object matching, mesh representation, computer animation, etc. A novel skeletonization algorithm to extract visually satisfactory skeleton from arbitrary 3D objects is proposed. The skeletonization can deal with arbitrary shape representations, such as polygonal models or parametric surfaces. In implementation, we choose triangulated models to represent 3D objects without losing generality. The skeletonization algorithm contains two major steps. First, all the 3D model faces are regarded as charged planes. We initiate seed points with negative charges from the 3D model vertices. These seed points are pushed by electric-static force and they finally converge to local minimum positions. The force model we use here is called "visible repulsive force", which is the sum of all repulsive forces derived from the visible charged planes. In other words, faces that are invisible from the seed point do not contribute in the force model. Then, these local minimum positions are connected to complete the skeleton. Every connection in the skeleton is determined according to the neighborhood relationship defined by the 3D model edge connectivity information. Currently, we have implemented an object deformation system and the generated skeletons are used to be the deformation reference lines.application/pdf982868 bytesapplication/pdfzh-TWotherhttp://ntur.lib.ntu.edu.tw/bitstream/246246/20060927122916679351/1/2003vrf.pdf