Iris-sensor authentication using camera PRNU fingerprints

Integrity and authenticity are important issues in biometrics security. Digital image forensics have shown that the integrity and authenticity of an image can be evidenced by a specific sensor fingerprint. If an image contains the sensor fingerprint it is assumed to be truly taken by that sensor. In such a way, an image can be authenticated by identifying it's sensor. However, this fingerprint may be forged. We calculated the photo response non uniformity (PRNU) as sensor fingerprint and estimated how well it differentiates iris scans of the CASIA V4 iris-sensors. The distinction rate varies largely between the sensors, with equal error rates between 0.21 and 23.26%. By inserting a fingerprint of one sensor into iris-scans of another sensor we successfully forged several iris scans. This forgery required estimating the best embedding-strength for the fingerprint, and we present a useful heuristic to do so. The triangle test is proposed as a countermeasure against this attack. However, it was shown not to be very effective. This could be explained by the fact that the quality of a sensor fingerprint determines if a forgery may be detectable or not by the triangle test. It is possible that the special content of the images limits the applicability of the PRNU, its forgery and the detection of the forgery in iris databases.