Mohammad Ramezani - Computer Vision Res. Lab, Electrical Engineering Faculty, Sahand University of Technology, Tabriz, Iran
Hossein Ebrahimnezhad - Computer Vision Res. Lab, Electrical Engineering Faculty, Sahand University of Technology, Tabriz, Iran

In this paper, we propose a 3D object indexing and retrieval system using elliptical descriptor for a large number of 2D silhouette views. Each silhouette is modeled by one ex-ellipsetree using Nelder-Mead optimization algorithm. The main advantage of using a large number of views is to eliminate therotation effect of 3D objects in space. By capturing many views from uniformly distributed points of view, most especial visual properties of a 3D model are used for retrieving assignment. The task of retrieving the 3D model is considered as the geometric analysis procedure and comparison between the query and the3D model is converted to compute the Euclidean distance of elliptical descriptor, which is previously extracted for eachmodel. The proposed 3D model retrieval algorithm has been evaluated on SHREC-W and McGill 3D shape benchmarks. Experimental result and comparison with other earlier approaches show the effectiveness of the proposed 3D model retrieval method.

3D Objects’ Retrieval; 2D Silhouette; Ex-Ellipse Fitting; Nelder-Mead Optimization Algorithm