Improving Face Recognition using DTW, PCA and Neural Network

 

 

 

Mohammed Kamel BENKADDOUR                           Abdenncer BOUNOUA

E-mail:   Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser.                                    Cette adresse e-mail est protégée contre les robots spammeurs. Vous devez activer le JavaScript pour la visualiser.

 

Laboratory Communication Network & Architecture Multimedia RCAM   DJILLALI LIABBES University Sidi Bel Abbes, Algeria

 

 

 

 

 

Abstract —  In this paper, we present a face recognition method combining PCA (principal Component Analysis), DWT (Discrete wavelet transform) and neural networks. This method consists of  four steps: i) Preprocessing, ii) Dimension reduction using DWT, iii)  feature extraction using PCA and  iv) classification using neural   network. To validate this work, we have tested this technique on frontal images of the ORL and Yale databases.

 

KeywordsFace Recognition, Biometrics, PCA, neural networks, feature extraction, eigenfaces , DTW.

 

I. INTRODUCTION

 

With  the   development  of   computers,  the   idea  of automatic  recognition  is  born,  it  is  the  beginning  of modern biometrics, it  has  gained considerable attention and many approaches have been developed and proposed by the scientific community using several methods based on fingerprints, iris, voice or face.

 

Facial  modality is  distinguished from  the  rest  of  the other  modalities  by  the  simplicity of  the  identification systems, and receives an increased attention because of its non-invasive nature, in the sense that it does not require the cooperation of the individual.

 

Before detailing the techniques used in this paper, we will   present   first   an   overview   of   studies   done   by researchers in cognition and facial recognition. The first work on face recognition began in the early 1970s [5]. But it’s just the last thirty years that research on face recognition has grown extensively and became of a great interest. For this reason, many several approaches have been proposed in the literature [3], they can mainly be classified into three groups:

 

-   Hybrid   approaches:   they   combine   both   types   of methods, potentially offering the best of both.

 

In this paper, first the PCA technique’s of how to extract the discriminate feature vectors and dimension reduction using DWT (Discrete wavelet transform )   is illustrated, and then, the practical adjustments necessary to implement the hybrid identification technique called Neural-PCA are outlined. After that, the experimental results obtained by each method by analyzing their performance, followed by a discussion with results interpretation are presented.

 

 

télécharger l'article