A Novel Approach to Improve Face Recognition Process Using Automatic Learning

A Novel Approach to Improve Face Recognition Process Using Automatic Learning

Yacine Gafour (EEDIS Laboratory, Djillali Liabes University, Sidi Bel Abbes, Algeria), Djamel Berrabah (EEDIS Laboratory, Djillali Liabes University, Sidi Bel Abbes, Algeria) and Abdelkader Gafour (EEDIS Laboratory, Djillali Liabes University, Sidi Bel Abbes, Algeria)
Copyright: © 2020 |Pages: 25
DOI: 10.4018/IJCVIP.2020010103
OnDemand PDF Download:
List Price: $37.50
10% Discount:-$3.75


In real-life applications, the appearance of a face changes significantly due to variations in expression, lighting, aging, exposure, and occlusion, which makes face recognition difficult. We present in this article a new approach for facial recognition. This approach is based on a set of variants of the Ho-LBP descriptor that we have proposed. In fact, the presentation of the images using a set of variants of the Ho-LBP descriptor helps the classifier to learn better. In addition, these variants are combined to improve the performance of facial recognition. We evaluated the effectiveness of our approach on ORL, Extended Yale B, and Feret databases. The obtained results are very promising, especially when compared with those of existing approaches. They show that our approach improves the accuracy of facial recognition in a very efficient way and in particular to the variations of the poses and the changes of the luminance.
Article Preview

1. Introduction

Biometric is a set of technologies, biometric technologies, that exploits human physical or behavioral characteristics such as fingerprint, signature, iris, voice, face and hand gestures to identify or recognize people. In terms of security and especially for authenticating people, biometrics offers more advantages than other existing methods such as keys, identification numbers (IDs), passwords and magnetic cards. For example, biometric technologies are safer than automatic access control cards and their secret codes which can be stolen or misplaced very easily.

Face recognition uses biometric settings to know or verify a person's identity. A face recognition system for security and surveillance must be robust, especially to posture variations and luminance changes. Facial recognition has attracted a great deal of attention and has been developed considerably in recent years. Although its performance is well achieved in a controlled environment, it is still far from satisfactory in real applications. There are many effective algorithms for solving facial recognition problems satisfactorily, but variations in expression, pose, occlusion and illumination are still critical issues that affect the performance of facial recognition. In fact, changing the pose and luminance of the same face can radically change the appearance of the person. So, these changes make the face recognition process difficult since they can affect the accuracy of this process.

Our main contribution in this paper is to develop a robust approach, especially with regard to the variation of poses and changes in luminance. In fact, we combine a set of variants of the Local Binary Pattern (LBP) (Ojala, Pietikäinen, & Harwood, 1996; Ojala, Pietikäinen, & Mäenpää, 2001) descriptor to improve accuracy and increase the chances of recognizing faces. To verify the performance of our approach, we used the three most used classifiers, which are: support vector machine (SVM), Random Forest (RF) and K-nearest neighbors (K-NN) algorithm. Then, we show the efficiency of the variants of our descriptor named Honeycomb-Local Binary Pattern (Ho-LBP) (Gafour, Berrabah, & Mahmoudi, 2018). It is a descriptor adapted to the honeycomb structure to describe a grayscale pixel in an image. This structure makes it possible to reduce the effects of the change of brightness and the variation of the poses to recognize the faces. The remainder of the paper is organized as follow: in section 2, we present the state of the art on descriptors (descriptors or LBP variants) and the field of face recognition. Then, in section 3, we explain the proposed approach for face recognition. Section 4 details the obtained results and discusses the performances of our approach within different datasets. Finally, we conclude this paper in Section 5.

Complete Article List

Search this Journal:
Volume 13: 1 Issue (2023): Forthcoming, Available for Pre-Order
Volume 12: 4 Issues (2022): 1 Released, 3 Forthcoming
Volume 11: 4 Issues (2021)
Volume 10: 4 Issues (2020)
Volume 9: 4 Issues (2019)
Volume 8: 4 Issues (2018)
Volume 7: 4 Issues (2017)
Volume 6: 2 Issues (2016)
Volume 5: 2 Issues (2015)
Volume 4: 2 Issues (2014)
Volume 3: 4 Issues (2013)
Volume 2: 4 Issues (2012)
Volume 1: 4 Issues (2011)
View Complete Journal Contents Listing