Do biometric images follow Benford's law?

Tampering of biometric samples is becoming an important security concern. Tampering can take place at the sensor level (spoofing), and through the backend, e.g., replacing the template with another sample. One example of backend attack is manipulating the original biometric image, i.e. contaminating the template. We study one particular aspect of tampering: image manipulation. In the forensics literature, Benford's law has been reported to be very effective in detecting tampering of natural images. In this paper, our motivation is to examine whether biometric images will follow the Benford's law and whether or not they can be used to detect potential malicious tampering of biometric images. We find that, the biometric samples do indeed follow the Benford's law; and the method can detect tampering effectively, with Equal Error Rate (EER) of 0.55% for single compressed face images, 2.7% for single compressed fingerprint images, 4.3% for double compressed face images and 3.7% for double compressed fingerprint images.