The Segmentation Of 3D Models And Its Applications

Multimedia has seen three waves so far: sound, images, and video. We are presently witnessing the arrival of the fourth wave of digital multimedia: Geometry and Digital Geometry Processing.3D model segmentation and shape description has become the most active research hotpot of Digital Geometric Process, and the key problem of parameterization, texture mapping, geometry morph etc. As a new framework of multimedia retrieval, based on local salient feature oriented segmentation, P3DS (Perceptual 3D Shape) has been accepted by MPEG-7, which is based on the theory of psychology, psychology physics and human vision. This thesis, in both the theoretical perspective and the practical perspective, probes into 3D model segmentation and shape description. A survey was given to summarize 3D mesh model segmentation techniques, including latest achievements, classification and application in this field. The main contributions of this thesis are summarized as follows:1) Proposed a novel method using principal curve to describe the shape feature of 3D models. After projecting the mesh into a spatial smooth cubic curve– principal curve, equally segmenting and sampling along the length of curve was implemented to get an ordered and uniform distribution of dataset. And then we matched the shape feature of one dimension manifold instead of that of 3D meshes, to overcome the difficulties of multi-resolution, local morph, and noise etc. Experiments prove that this shape descriptor is invariant under translation, rotation and scaling. Some existent problems and prospects of this technique were also discussed.2) Proposed a novel method using KS pincipal curve to merge unregistered, noisy data set of Human Body Scan and Measurement. And the crossing zero of curvature was used to segment and estimate the target feature points from the KS principal curve. An optimization scheme named variable fuzzy partition was also presented to improve the searching for local feature area.3) A novel implementation of Dijkstra algorithm based on CSR data structure of 3D mesh model with sparse matrix adjacency was presented too. The intermediate result was reused to save the computing resource and improve the speed. This algorithm provided a geodesic distance for the cluster segmentation of 3D mesh model.