Comparative analysis of score level fusion techniques in multi-biometric system

Abstract

Multimodal biometric systems have garnered significant interest from researchers owing to their applicability in security and access control. Despite the development of numerous score level fusion techniques for multimodal biometrics, most of them have concentrated solely on enhancing fusion accuracy, neglecting the potential advantages of various score level techniques. This research investigates the comparative performance of four different score level fusion approaches for multimodal recognition of combined face and fingerprints biometrics: Product rule, Weighted Sum rule, Simple Sum rule, and Max rule method. Five hundred and seventy (570) sample images from 190 students of Ladoke Akintola University of Technology (LAUTECH), Ogbomoso, used in this study were acquired using CMITech camera for faces and digital personnel for fingerprints, respectively. The images consist of three (3) samples of each biometric trait. Three hundred and forty-two (342) images of these traits were used for training while two hundred and twenty-eight (228) images were used for testing. The acquired images were pre-processed using histogram equalization, features extraction was done using Principal Component Analysis. Euclidian distance and Manhattan distance was used for generating the matching score of face and fingerprint feature respectively while Min-max was used to normalize each score. The fused score of each technique was used for identification. The results obtained was evaluated using False Acceptance Rate (FAR), False Rejection Rate (FRR) and Recognition Accuracy (RA) and Recognition Time (RT). Experimental results revealed that the Weighted Sum Rule outperformed other techniques, achieving a FAR of 1.75%, FRR of 5.85%, RA of 95.18%, and RT of 56.12 seconds. Comparatively, the Product Rule, Simple Sum Rule, and Max Rule demonstrated lower performance metrics. This study underscores the efficacy of the Weighted Sum Rule as a superior score-level fusion technique for developing advanced multimodal biometric systems, particularly in applications requiring high security and reliability.