A new fast method for 3D indexing and retrieval based on mapping techniques

Abstract

In this paper we present a new fast method for 3D objects indexing and retrieval based on two-dimensional (2-D) views. The set of views are automatically generated around OX, OY and OZ by rotating each 3D object of database through 360° with taking views at pose intervals of 45°, given 24 images per object. The obtained views are afterward described by two descriptors: Angular Radial transform adopted by MPEG-7 and Fourier descriptor. The similarity of 3D objects is calculated as a linear combination of euclidian metrics. In order to lower the cost of search, we propose to apply a mapping from metric space to vector space of 3D descriptors by using a contractive function. We have implemented an application for indexing and retrieval of 3D objects by using Princeton 3D Shape Benchmark database. The performance of our retrieval system was measured in terms of recall and precision and Computation efficiency. The obtained results prove the effectiveness of proposed method and its superiority over automatic selection based on Curvature Scale Space.