Cancelable Biometrics based on Biometric Salting

Abstract

Nowadays biometrics systems for identification or authentication of a person are everywhere. These system have a number of advantages. In particular, biometrics traits cannot be lost or forgotten compared to passwords. Moreover biometric identification offers good accuracy. However, their uses raises several privacy concerns, especially in their storage. In fact, if a password is stolen, it can be replaced by a new password. This is not possible in biometrics. To overcome the security problems, biometric cryptosystems (BCS) and cancelable biometrics (CB) represent emerging technologies of biometrics template protection addressing this concerns and improving public confidence and acceptance of biometrics. BCS are designed to securely bind a digital key to a biometric or generate a digital key from a biometric offering solutions to biometric-dependent key-release and biometric template protection while CB consist of intentional, repeatable distortions of biometric signals based on transforms which provide a comparison templates in the transformed domain.<br/> <br/> In this dissertation, a cancelable biometric scheme for iris recognition system is proposed. The first proposed CB method uses the reduced random permutation and binary salting (RRP-BS). RRP-BS consists of random permutation of binary iris template followed by the orthogonal binary salting. The random permutation perturbs the rows of iris template structure and eliminates some rows of iris template. This guarantees the non-invertibility of CB scheme even though the all of bio-security keys is stolen. Then this CB scheme also proposes an orthogonal binary salting method, where the random binary keys are generated by Gram-Schmidt orthogonalization. The orthogonality of random keys maximizes the Hamming distances among binary-salted templates. Thus, the inter classes (different users) are discriminated while the intra class (one user) is well identified. While this method has good performance and unlinkability, its non-invertibility is vulnerable to muliplicity or hill-climbing attacks. The second proposed method uses more robust non-invertibility transform based on the first method. We use the RRP-BS as the biometric salting, and use the Hadamard product for enhancing the non-invertibility of salted data. Moreover, to overcome the shortcomings of perserving the keys of the conventional salting methods, we generate several templates for an input, and define non-coherent and coherent matching regions among these templates. We show that salting the non-coherent matching regions is less influential on the overall performance. Specifically, embedding the noise in this region does not affect the performance, while making the data difficult to be inverted to the original. For the evaluation, we use three datasets, namely CASIA V3 iris-interval, IIT Delhi iris, and ND-Iris-0405. The extensive evaluations show that the proposed algorithm yields low error rates and good intra/inter classification performances, which is better or comparable to the existing methods. Moreover, the security analysis ensures that the proposed algorithm satisfies non-invertibility and unlinkability, and is robust against several attacks as well.