AI for Biometric Authentication Systems

Tim Van hamme, Giuseppe Garofalo, Sander Joos, Davy Preuveneers, Wouter Joosen

Abstract: Biometric authentication is currently one of the most attractive authentication methodologies, as it is very user friendly and is the most secure option for consumer grade mobile device authentication. Furthermore, biometric authentication can provide unique security properties, as it is the only authentication factor that is hard to share or steal. Machine learning has further advanced biometric authentication performance and helps detect well known attack vectors such as spoofing attacks. Sadly, the use of machine learning enlarges the already elaborate attack surface of biometric authentication systems. We further explore the how, pros and cons of machine learning for – the interesting security problem – biometric authentication.