Hybrid Level Fusion Schemes for Multimodal Biometric Authentication System Based on Matcher Performance

Abstract

The performance of the multimodal system was improved by integrating both the physiological and behavioral characteristics of an individual. Usually, the fusion is carried out either in the score or feature level. Multimodal biometric system against the unimodal system is explained below. This system was considered in order to overcome several demerits that were found in the former system. The overall recognition rate of the biometric system was improved by implementing the multimodal systems. ECG, face, and fingerprint were integrated in the new level of fusion named hybrid fusion scheme. In the hybrid fusion scheme, the scores from the feature level fusion were fused along with the best unimodal system (ECG) by using score level fusion techniques. Feature vectors were obtained by processing the signal as well as the images obtained from the databases FVC2002/2004, Face94, and PhysioNet (MIT-BIH Arrythmia) after the process of feature extraction. Matching scores and individual accuracy were computed separately on each biometric trait. Since the matchers on these three biometric traits gave different values, matcher performance-based fusion technique was proposed on the specified traits. The two-level fusion scheme (score and feature) was carried out separately, for analyzing their performances with the hybrid scheme. The normalization of the scores was done by using overlap extrema-based min–max (OVEBAMM) technique. Further, the proposed technique was compared with Z-Score, tanh, and min–max by considering the same traits. The performance analysis of these traits with both unimodal and multimodal systems was done, and they were plotted using receiver operating characteristic (ROC) curve. The proposed hybrid fusion scheme has leveraged the best TPR, FPR, and EER rates as 0.99, 0.1, and 0.5, respectively, by using the normalization techniques with the weighted techniques like confidence-based weighting (CBW) method and mean extrema-based confidence weighting (MEBCW) method.KeywordsMultimodal biometric systemMatcher performanceHybrid level fusionWeighted methods