Cryptanalysis of the Fuzzy Vault for Fingerprints: Vulnerabilities and Countermeasures

Abstract

The fuzzy fingerprint vault is a popular approach to protect a fingerprint's minutiae as a building block of a security application. In this thesis simulations of several attack scenarios are conducted against implementations of the fuzzy fingerprint vault from the literature. Our investigations clearly confirm that the weakest link in the fuzzy fingerprint vault is its high vulnerability to false-accept attacks. Therefore, multi-finger or even multi-biometric cryptosystems should be conceived. But there remains a risk that cannot be resolved by using more biometric information of an individual if features are protected using a traditional fuzzy vault construction: The correlation attack remains a weakness of such constructions. It is known that quantizing minutiae to a rigid system while filling the whole space with chaff makes correlation obsolete. Based on this approach, we propose an implementation. If parameters were adopted from a traditional fuzzy fingerprint vault implementation, we would experience a significant loss in authentication performance. Therefore, we perform a training to determine reasonable parameters for our implementation. Furthermore, to make authentication practical, the decoding procedure is proposed to be randomized. By running a performance evaluation on a dataset generally used, we find that achieving resistance against the correlation attack does not have to be at the cost of authentication performance. Finally, we conclude that fuzzy vault remains a possible construction for helping in solving the challenging task of implementing a cryptographically secure multi-biometric cryptosystem in future.