Eye movement-driven defense against iris print-attacks

This paper proposes a methodology for the utilization of eye movement cues for the task of iris print-attack detection. We investigate the fundamental distortions arising in the eye movement signal during an iris print-attack, due to the structural and functional discrepancies between a paper-printed iris and a natural eye iris. The performed experiments involve the execution of practical print-attacks against an eye-tracking device, and the collection of the resulting eye movement signals. The developed methodology for the detection of print-attack signal distortions is evaluated on a large database collected from 200 subjects, which contains both the real ('live') eye movement signals and the print-attack ('spoof') eye movement signals. The suggested methodology provides a sufficiently high detection performance, with a maximum average classification rate (ACR) of 96.5% and a minimum equal error rate (EER) of 3.4%. Due to the hardware similarities between eye tracking and iris capturing systems, we hypothesize that the proposed methodology can be adopted into the existing iris recognition systems with minimal cost. To further support this hypothesis we experimentally investigate the robustness of our scheme by simulating conditions of reduced sampling resolution (temporal and spatial), and of limited duration of the eye movement signals. Eye movement cues are used for iris print-attack detection.Feature extraction is based on statistical modeling of signal distortions.Print-attack detection performance reaches a top rate of 96.5%.Extended analysis regarding the signal recording conditions is performed.