A COMPARATIVE STUDY OF HUMAN FACES RECOGNITION USING PRINCIPLE COMPONENTS ANALYSIS AND LINEAR DISCRIMINANT ANALYSIS TECHNIQUES

Abstract

This paper presents a comparative study of human faces recognition using two feature extraction techniques: Principle Components Analysis (PCA), and Linear Discriminant Analysis (LDA). The performance of these techniques is evaluated and compared to find the best technique for human faces recognition. The experiments are carried out on the Olivetti and Oracle Research Laboratory (ORL), University of Manchester Institute of Science and Technology (UMIST), and Japanese Female Facial Expression (JAFFE) face databases, which include variability in affectation, facial details, and expressions. The obtained results for the two techniques have been compared by varying the train images/test images ratio on three levels: 80/20, 60/40, and 40/60. The experimental results show that the LDA feature extraction technique gives better performance than the PCA technique. The highest recognition rate is recorded for the LDA technique (recognition rate=95.981%) when the train images/test images ratio is (80/20). On the other side, the highest recognition rate that is recorded for the PCA technique is 94.027% when the train images/test images ratio is (80/20). The PCA and LDA techniques are implemented and their performance is measured using the MATLAB (2013) program.