Adaptive SVM fusion for robust multi-biometrics verification with missing data

Abstract

Conventional multimodal biometrics systems usually do not account for missing data (missing modalities or incomplete score lists) that is commonly encountered in real applications. The presence of missing data in multimodal biometric systems can be inconvenient to the client, as the system will reject the submitted biometric data and request for another trial. In such cases, robust multimodal biometric verification is needed. In this paper, we present the criteria, fusion method and performance metrics of a robust multimodal biometrics verification system that verifies the client's identity at any condition of data missing. A novel adaptive SVM classification method is proposed for missing dimensional values, which can handle the missing data in multimodal biometrics. We show that robust multibiometrics imposes additional requirements on multimodal fusion when compared to conventional multibiometrics. We also argue that the usual performance metrics of false accept and false reject rates are insufficient yardsticks for robust verification and propose new metrics against which we benchmark our system.