Producing secure multimodal biometric descriptors using artificial neural networks

Abstract

With the rapidly growing use of biometric authentication systems, the security of these systems and the privacy of users have attracted significant attention in recent years. Multi-modal biometrics have been able to improve the accuracy of the system but require additional bandwidth to exchange the data. Fragile watermarking has been used to allow the transmission of both biometric templates using the amount of data required to transmit one of them, that is, the cover image, while securing these templates against attacks. Despite the high accuracy of these systems, communicating such templates imposes risks towards the privacy of the users. In this study, a new method is proposed to generate fixed-size descriptors for the face and fingerprint templates, including the timestamp of the transmission and a unique system identifier. The inclusion of the timestamp enables the system to detect and deny replay attacks, while the unique system identifier maintains the privacy of the users. The experiments conducted to evaluate the proposed method have shown that the proposed method has been able to achieve these features while maintaining high recognition rates, 99.41% and 99.32%, similar to the use of the entire biometric templates in the matching stage.