Score-level fusion in multiple biometrics using non-linear classification

Abstract

This paper proposes a multiple biometric system using non-linear classifiers instead of fusion functions such as weighted sum [1]. In the proposed system, multiple matching scores from individual biometric systems are considered as a score vector which is classified by Support Vector Machine (SVM), Kernel Fisher Discriminant (KFD) and Bayesian Classifier. Experiments have been conducted on Set 3 of NIST BSSR1 (Biometric Scores Set - Release1) data, and the performance of classifiers is evaluated in terms of FAR (False Accept Rate), FRR (False Reject Rate), HTER (Half Total Error Rate) and the ROC (Receiver Operating Characteristic) curves. The experimental results demonstrate that multiple biometric systems using non-linear classification methods provide higher verification performance than single biometric systems.