Face Recognition using PCA versus ICA versus LDA cascaded with the Neural Classifier of Concurrent Self-Organizing Maps Victor-Emil Neagoe, Senior Member IEEE, Alexandru-Cristian Mugioiu Dept. Electronics, Telecomm. and Inform. Technol. “Politehnica” University of Bucharest Bucharest, Romania E-mail: victoremil@gmail.com Ioan-Anton Stanculescu School of Informatics University of Edinburgh Edinburgh, United Kingdom E-mail: i.a.stanculescu@sms.ed.ac.uk Abstract—We present a comparison of three feature selection methods for face recognition: Principal Component Analysis (PCA), Independent Component Analysis (ICA), and Linear Discriminant Analysis (LDA). One also considers evaluation of the neural classifier based on Concurrent Self-Organizing Maps, (CSOM), previously introduced by first author of this paper. The ORL Database of Faces is used for experiments and the corresponding results of evaluation are given. Keywords-face recognition, PCA, ICA, LDA, SOM, CSOM I. INTRODUCTION The September 11 and the July 7 terrorist attacks have changed the way, the world looks towards security. Hence, the need for a robust and effective biometric system for security application has been highlighted by security agencies all over the world. Biometrics was traditionally defined as the study of measurable biological characteristics. However, in computer vision, biometrics refers to a measurable physical or behavioral characteristic used to recognize the identity, or verify the claimed identity of a person through automated means. Although extremely accurate and reliable biometric methods of personal identification exist (e.g., fingerprint analysis, iris scan), these techniques rely on the cooperation of the participants. Automatic face recognition is a challenging problem which has received much attention during the recent years. The advantage of face recognition is that it is a non-intrusive technique that can be effective without participant’s cooperation or knowledge; this makes it especially suitable for surveillance purposes. Face recognition systems have a large variety of applications like: access control, electronic transactions, identity authentication, hidden surveillance, law enforcements and a lot more. Face recognition can be divided into two basic applications: identification and authentification. In the identification problem, the face to be recognized is unknown and is matched against faces of a data base containing known individuals. In the authentification (verification) problem the system confirms or rejects the claimed identity of the input face. This paper will address the general problem of face recognition and no particular distinction will be made among the two problems as the challenges and the used techniques are basically the same. Figure 1. Comparative market share by technology type for biometric industry in the interval 2006 – 2011. The world biometric industry market for face recognition is predicted to increase with 100% from 2008 to 2011. Face recognition as a particular case of pattern recognition contains two processing stages: feature selection and classification (Fig. 2). Figure 2. Comparative market share by technology type for biometric industry in the interval 2006 – 2011. This paper presents the results of experiments and evaluation of the cascade where the first processing stage is one of the following feature selection algorithms: Principal Component Analysis (PCA), Independent Component Analysis 978-1-4244-6363-3/10/$26.00 c 2010 IEEE 225