Face Recognition Using Principle Component Analysis And Linear Discriminant Analysis

Abstract

Now day face recognition is one of the most dominant biometrics used in the area of information security, law enforcement, access control, surveillance and multimedia search engine. The non-intrusive nature of this technology makes more advantage than other biometric technologies. In the last few years, a number of researcheson face detection and recognition have been conducted by researchers in different fields claimed to have satisfactory results in the preceding researches. However, until now there is no a single face detection and recognition method that satisfy in all circumstances. What makes to be unsatisfactory is that it could be affected by external and internal factors such as light variation, expression variation, angle variation, image background, age variation, and motion. These variations can be reduced by applying various image preprocessing techniques. This thesis proposes and describes a research attempt at designing and developing a prototype face detection and recognition system. The system is developed using the Java programing language with the help of OpenCV 3.1.0 image processing library. For face detection purpose, we have been used the Viola and Jones method and its performance was evaluated in both real-time and non-real-time environment. The accuracy of the detector is tested using images from Carnegie Mellon University and Massachusetts Institute of Technology(CMU-MIT), Labeled Faces in the Wild (LFW), image from Faculty of Industrial Engineering (FEI) and self-preparedface databases and produces agood result. The study also considers various image preprocessing techniques such as histogram equalization, Bilateral Filter, face masking, and image resizing to increase the accuracy the recognizer. The principal component analysis (PCA) and linear discriminant analysis (LDA)dimension reduction techniques were used to extract image information and classified the image based on distance measurement techniques. The study also conducts comparative analyses of these two methods to determine their relative strengths, weaknesses,and suitability for face recognition. The experiment resultsof masked FEI, ORL and Self-prepared datasets produce an accuracy rate of 95%, 91%, and 91.25% respectively using PCA method. And experiment result of LDA method produce accuracy rate of 96%, 86.25%, 83.75% for the three different database using masking techniques.The results areencouraging and with more optimization works better results can be achieved