Generalized Benford’s Law for Fake Fingerprint Detection

Human fingerprints are considered to be the most extensively used modality in state-of-the-art biometric based security systems. A newly developed potential threat occurs in the form of specialized fake fingerprint, known as the double-identity fingerprint which is created by careful alignment of two genuine fingerprints to ensure smooth ridges at the intersection points and with high chance of being matched by the criminal as well as accomplice. In this paper, we propose a countermeasure to resolve this problem with generalized Benford's law in conjunction with support vector machine classification. The generalized Benford's law is found to validate the probability distribution of first nonzero digits from block-DCT coefficients of genuine fingerprint images. The experimental results support that the synthesized fake double-identity fingerprints do not inherently follow the generalized Benford's law and hence the proposed method can be effective to discriminate from the genuine class with high accuracy.