Local Structure Based Fingerprint Authentication Systems with Template Protection

Abstract

Fingerprint authentication systems are widely used in military and civil applications due to the distinctiveness and stability that fingerprints can supply. Research in fingerprint authentication systems has made great achievements in the last several decades with authentication technologies in the unencrypted domain becoming mature. However, biometric templates themselves also need to be protected. Conventional encryption technology will lead to poor performance due to the nature of biometric uncertainty presented in each fingerprint image. There are following two streams of research, bio-cryptosystem and cancellable biometrics, in replacing conventional encryption technology. Specifically, bio-cryptosystem is an emerging biometric-key framework that can protect both biometric templates and cryptographic key templates, while cancellable biometrics transforms raw features into a different domain so that the compromised template in the transformed domain cannot reveal the raw template. However, two issues have not been resolved. First, existing fingerprint based bio-cryptosystems and cancellable biometric systems cannot achieve satisfactory authentication performance because it is difficult to overcome fingerprint uncertainty in the encrypted domain. Second, the existing cancellable biometric systems suffer from attacks via record multiplicity (ARM). In this thesis, we tackle the above two critical biometric authentication issues by designing the fingerprint authentication systems based on some stable local structures in the encrypted domain. The work that has been pursued in this thesis is briefly described below: Firstly, an alignment-free bio-cryptosystem based on modified Voronoi neighbor structures (VNSs) is proposed. The proposed method obviates fingerprint pre-alignment by utilizing the rotation- and translation-invariant feature representations extracted from modified VNSs. Fingerprint uncertainty is mitigated by the construction of VNSs and modification of the originally formed VNSs. Experimental results using the publicly-available databases show the validity of the proposed scheme. Secondly, we propose to adopt the Delaunay quadrangle-based structure to address the local structural change under non-linear distortion that the Delaunay triangle-based structure experiences. A unique topology code originated from each Delaunay quadrangle can assist in accomplishing good local registration in the presence of nonlinear distortion and can further enhance the security of the overall system on top of what is provided by the secure sketch. Experimental results and security analysis show that the Delaunay quadrangle-based system with topology code can achieve better performance and higher security level than the Delaunay triangle-based system, the Delaunay quadrangle-based system without topology code and some other similar systems. Finally, we develop a dual-structural-layered fingerprint cancellable template system. The new cancellable template can mitigate the negative impact of non-linear distortion on those minutiae that are far away from the reference point. Furthermore, the proposed method increases the difficulty of launching the ARM through hiding the transformation parameter by using some stable local features, which is a clear advantage over those existing cancellable templates.