Security Analysis of Alignment-Robust Cancelable Biometric Scheme for Iris Verification

Abstract

In cancelable biometric (CB) schemes, secure biometric templates are generated by applying, mainly non-linear, transformations to the origin data. The cancelable templates should satisfy the requirements of irreversibility, unlinkability, and revocability with high accuracy. However, existing cancelable biometric schemes have been demonstrated that their security is overestimated. Many well-known cancelable biometric schemes have been proven vulnerable to some attack models. In this paper, we analyze a recent alignment-robust cancelable biometric scheme called Random Augmented Histogram of Gradients (R HoG) that is not as unlinkable as proposed. Moreover, we propose two schemes to attack the unlinkability of R HoG. One is that two cancelable templates from different applications are directly connected according to the leaked tokens, and the other is based on the reverse of Z-score transformation, which can achieve higher linkability. Experimental results on CASIA-IrisV3-Interval show that the cancelable biometric template generated by R HoG has high linkability with a maximum link success rate of 95.62%.