ISSN 2079-3537      

 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                                                                                             





Scientific Visualization, 2018, volume 10, number 2, pages 84 - 94, DOI: 10.26583/sv.10.2.07

New Approach for 3D Mesh Retrieval Using Artificial Neural Network and Histogram of Features

Authors: M. Bouksim A,1, K. Arhid A,2, F.R. Zakani A,3, M. Aboulfatah B,4, T. Gadi A,5

A Laboratory of Informatics, Imaging, and Modeling of Complex Systems (LIIMSC)

Faculty of Sciences and Techniques, Hassan 1st University, Settat, Morocco

B Laboratory of Analysis of Systems and Treatment of Information (LASTI)

Faculty of Sciences and Techniques, Hassan 1st University, Settat, Morocco

1 ORCID: 0000-0001-5509-2529, mbouksim@gmail.com

2 ORCID: 0000-0001-8850-1106, khadija.arhid@gmail.com

3 ORCID: 0000-0002-4035-559X, fatima.rafiizakani@gmail.com

4 ORCID: 0000-0002-6382-7860, mohamed.aboulfatah@uhp.ac.ma

5 ORCID: 0000-0002-2174-5816, gtaoufiq@yahoo.fr

 

Abstract

Recently 3D models become more popular in diverse fields such as medicine biology and engineering; this expansion created the need of a robust descriptor system that will allow a fast and compact classification, comparison and retrieval of 3D models. A diversity of methods and approaches have been proposed to solve this problem, but recently researchers got interested in the use of the potential and the effectiveness of machine learning methods to create a powerful retrieval system. In this paper, we present a new method to extract a descriptor or signature representing the 3D model. The proposed method consists of using an artificial neural network (ANN) trained with a histogram of features extracted directly from the 3D object; this last point helps to train the ANN fast and with consistent data. Once trained we concatenate the result of the hidden layers to be used as a descriptor in the retrieval system. The achieved experimental demonstrate the power and the effectiveness of our method which outperform some well-known methods in the literature.

 

Keywords: 3D model, 3D object retrieval, 3D shape retrieval, 3D shape matching, Artificial neural network, 3D object signature.