Biometric system : face recognition

Abstract

In the last decades, there has been a stronger focus on security around the world. One of the important issues in security is the need of correctly authenticate a person. Traditional methods of establishing a person's identity include passwords, keys or identification cards, but these surrogate representations of identity can easily be lost, shared, manipulated or stolen thereby compromising the intended security. Biometrics provides better security, higher efficiency, and increased user convenience. Among the more acceptable features used in biometrics is the face which provides a more direct, friendly and convenient identification method. Facial research in computer vision can be divided into several areas, such as face recognition, face detection, facial expressions analysis. In our thesis we are interesting on face recognition since it is a biometric method which has become more relevant and needed. Many methods of face recognition have been proposed. Basically, they can be divided into three categories: Global-Appearance-based methods, local appearance-based methods and hybrid methods. The most successful and well-studied techniques are the appearance-based methods. So, in this thesis some of them are implemented namely: Principle Component Analysis (PCA), Fisher Linear Discriminant (FLD), Singular Value Decomposition (SVD), and Discrete Wavelet Transform (DWT) or Wavelet Packet Decomposition (WPD). The local appearance-based methods can be divided into two groups: The ones that require the use of specific regions and the ones that simply partition the input face image into blocks without considering any specific regions. For the latter type we have implemented the Discrete Cosine Transform (DCT) method. Furthermore, we have applied DWT and WPD as preprocessing step for PCA, as well DWT for FLD and SVD in order to improve their performances. Then, we have proposed a new method which combines PCA and WPD in YCbCr color space called (YCbCr-WPD-PCA) L199]. In addition, we have introduced the multibiometrics system particularly multi-algorithm systems. In this thesis we have utilized multi-algorithm systems that consolidate the output of multiple feature extraction algorithms at score levels. We have used the fusion of two types and three types of feature vectors using four fusion rules i.e. minimum, maximum, mean, and product. Finally, many experiments were conducted using different databases namely YALE, ORL, and FEI databases. The accuracy of each method has been identified and a comparison was performed between them in terms of recognition rates, Equal Error Rates (ERR) or the Receiver Operating Curves (ROC)