Shape mining in three dimensional models databases.

摘要

Three-dimensional Shape Search is an emerging research area. This dissertation presents new algorithms for mining the shape information in a CAD model database. Mining for shape in a CAD database involves a shape representation, a similarity definition, and applications for using the shape information. This dissertation presents original contribution in (a) developing a new shape representation, (b) proposing a flexible similarity model for shape matching using the new representation and (c) demonstrating a visualization tool for browsing the models in a CAD database. The proposed shape representation is based on a volumetric decomposition of a mesh with swept components using the underlying sweep process. The key to identifying the different segments based on the inherent sweep process is to identify the characteristic cross-sections used to create them. An algorithm for analyzing the surface mesh using a sectional gauss image for obtaining these characteristic cross sections is presented. A segment graph formulation for representing the segmented objects has been presented. Further, a flexible similarity model has been proposed by posing the shape-matching problem as a graph-matching problem on their segment graphs. An edit-distance based inexact graph matching technique was implemented.;The flexibility in the similarity definition is illustrated with four sample schemes for the cost formulation in the edit distance computation. The demonstrated usecases are an exact shape search, parametric shape search, deformed shape search and partial shape search. In addition, a novel navigation paradigm for browsing the CAD models in a repository has also been presented. The proposed visualization using this paradigm is a three-dimensional graphical user interface displaying the objects in the database in a three-dimensional space. User testing of the proposed approach has been done and the results discussed.