Wolf Attack: Algorithmic Vulnerability in Biometric Authentication Systems

Abstract

Biometrics is a key technology for the authentication of individuals. As far as malicious impersonation attempt cannot be put an end, security is essential for biometric authentication systems. The security of biometric authentication systems is conventionally measured by false acceptance rate (FAR), the average probability of falsely accepting an individual claiming different identity. Since malicious attackers are not limited to use their own biometric features, clearly, FAR is not sufficient for the measure of security. On the other hand, it is pointed out that deep analysis of biometric authentication algorithms often shows that some irregular artificial patterns give very high acceptance rate against large portion of registered human templates. This is called wolf attack. Including this class of attack, in this paper, we formulate the security of biometric authentication systems, and demonstrate our recent results in fingerprint recognition systems. Finally, we introduce the theoretical framework for securing biometric authentication systems.