Class discriminating features-based SVM for face membership authentication

Face membership authentication is to decide whether an incoming face is that of an enrolled member or not. Face membership authentication is basically a two class (enrolled or unenrolled) classification where SVM (Support Vector Machine) has been successfully applied and shows better performance compared to the conventional threshold rule-based authentication methods. Most of previous SVMs for face membership authentication have been trained using image feature vectors extracted from face images in the training set. However, image features extracted from images are not robust to variations of illuminations, poses, and facial expressions. Moreover, enrolled/unenrolled class can be dynamically changing due to memberspsila secession or joining. And, unenrolled class is prohibitively huge. In this paper, we propose an effective class discriminating feature-based SVM for face membership authentication. The adopted features for training and testing the SVM is not extracted from face images but reflect discrimination between enrolled class and unenrolled class. Thus, the proposed SVM is relatively independent from variations in face images and less affected by changes in membership configuration. Through experiments, it is shown that the face membership authentication method based on the proposed SVM performs better than the threshold rule-based or the conventional SVM-based authentication methods and is relatively less affected by change in membership configuration.