Performance evaluation of face verification: a comparative study on different classifiers

Abstract

The task offace verification is to verify the identity or decide whether the a priori user is an impostor or not from the known a priori identity of the user. The paper presents the performance evaluation carried out using different classifiers for face verification. The paper initially describes the approaches used for the face representation. and classification of face verification system. It then evaluates the performance of the system by applying three types of classifier: template-based matching, artificial neural network classifier. and Bayesian classifier based on AT & T and local face daJasets. The measures used for performance evaluation are the false acceptance rate (FAR) and false rejection rate (FAR). Based on the experimental results, the artificial neural network classifier provides promising results for face verification with FAR of 4.44% and FRR 4.50% using AT&T face daJaset, and FAR of 3.88 and FRR 4.00 % using local face dataset.