Sppof-proofing fingerprint systems using evolutionary time-delay neural networks

With the new wave of affordable, small, and easy to use scanners, fingerprint-based biometric systems have been receiving an increasing attention. However, a major security concern is the possibility of intrusion by presenting a nonliving finger, be it a duplicate or a severed finger, to an electronic fingerprint scanner in order to gain access to a protected entity. It has been shown that one can spoof fingerprint scanners even with play-dohreg or gummy fingers. In order to circumvent this problem, one can read signals from the presented finger to verify its liveness and thus eliminate the threat of synthesized or cadaver finger attach. Earlier research has shown that the process of perspiration on the live fingertip skin presents a specific time progression that cannot be seen in cadaver or synthetic fingerprint scans, and thus phenomenon can be used as a measure of fingerprint liveness. However, the perspiration process demonstrates itself differently on different scanning technologies and thus a scanner-specific approach is needed. In this paper a new general evolutionary temporal neural network (GETnet) for perspiration-based liveness detection is proposed. It is shown that GETnet can arrive at a succinct solution that performs both feature extraction and classification on the raw fingerprint ridge signals. With the given variety of fingerprint scanners as well as the diversity of their operating conditions, including climate and user demographics, it is more efficient to automatically breed customized solutions for the perspiration-based fingerprint liveness detection through a general framework such as GETnet instead of tailoring feature extractors and classifiers to each and every different scenario