Fingerprint Template Protection using Compact Minutiae Patterns

Abstract

Due to their superior reliability in user authentication compared to traditional security tokens, biometrics are at the forefront of identity management systems. Despite their potential security benefits, the proliferation of biometric technologies is threatened by the public‟s association of the use of biometrics with an impending Big Brother scenario. Several vulnerabilities of the biometric recognition system remain to be dealt with before the public can be assured that their privacy will not be violated upon surrendering their biometric identifiers to the recognition system. The most serious breach of a biometric system is theft of the templates stored in its database. This problem is especially critical for fingerprints, whose reputation as the most mature and popular biometric modality is responsible for an explosive growth in fingerprint databases. Since a fingerprint is permanent throughout an individual‟s lifetime, and each person has only a limited number of fingerprints, the compromise of a fingerprint would result in a lifelong threat to the victim‟s security and privacy. To prevent the theft or misuse of fingerprint data, and consequently maintain the privacy of fingerprint technology users, fingerprint templates must be effectively secured during storage in a database. The relative infancy of this field of work and the challenging nature of this task means that there does not yet exist an agreed-upon solution to the problem of secure fingerprint template storage. The difficulty of this problem, combined with the urgency of developing an effective solution, has motivated the research presented in this thesis. The main objective of this thesis is to develop a novel mechanism for securing fingerprint templates during storage in a database. This objective has been realised via the proposal of a new fingerprint construct, which is a non-invertible fingerprint template protection scheme by its very nature. The crux of our scheme entails the representation of a fingerprint by a single N-node Pattern constructed using a small subset of N minutiae from the corresponding minutiae template. The sparsity of the resulting Pattern makes it impossible to reconstruct the original fingerprint template and it ensures that the Pattern is cancellable in the event of compromise. Furthermore, despite its sparsity, an N-node Pattern is found to have acceptable recognition accuracy in the cooperative-user scenario for which it is intended. The most important aspect of this fingerprint template protection scheme, which sets it apart from other methods in the associated literature, is that it incorporates only a small fraction of the entire fingerprint template in the generation of the protected template. Consequently, the protected template is intuitively more secure. The majority of this thesis is dedicated to a rigorous analysis of the proposed fingerprint construct. In summary, the analysis shows that our new construct is able to comply satisfactorily with the four properties of an ideal fingerprint template protection scheme: non-invertibility, cancellability, diversity, and performance. It thus seems worthwhile to further mine the potential of this promising new fingerprint template protection scheme. Other contributions in this thesis include an investigation into the intra-class variance of fingerprints acquired from cooperative users, two thorough literature reviews on existing fingerprint template protection schemes, the proposal of a new fingerprint alignment method, and a dissection of the methods used to implement fingerprint-based fuzzy vaults in the literature.