Program for Biometric Authentication based on Face Recognition

Face recognition is a special case of visual object recognition that presents a challenge with potentially high rewards in the field of computer vision and biometric authentication, and as such has over the last several years been in the focus of attention of security experts and researchers in various other domains. The main tasks addressed by face recognition are identification and verification, which are commonly solved by statistical pattern classification algorithms, called classifiers. The growing interest in the biometrics has significantly increased the number of available classifiers which solve the recognition problem in general, and identification task in particular. In order to evaluate the performance of different classification algorithms, common performance metrics are used. This thesis is aimed at determining the ways of improving a particular face identification system (based on OpenBR framework) in the company by reducing rate of type I and II errors of its classifier. These errors show ratio of falsely predicted and falsely unrecognized images to all, which have been subjected to classification, and are crucial in determining the level of performance of the authentication system. In this thesis an analysis of the existing company’s identification system is provided and available approaches to face recognition task are reviewed. Furthermore, binary and multiclass classification models with their respective performance metrics are described. Moreover, holistic-based classifiers and image dataset for system’s training and testing are prepared. Finally, ways of improving the face recognition system under consideration are proposed and tried with the help of the developed face identification system prototype. The results obtained from the tests indicate that implementation of the created AI-based classifier into the company’s authentication system should improve its performance and reduce the number of classification errors.