On the efficiency of the transformation-based template protection algorithms

Abstract

With the intense proliferation of biometric identification systems (BIS), several security con cerns about the vulnerability of the user templates have emerged. Indeed, unlike the conven tional security systems that are based on passwords or tokens which are renewable, a biometric template is not renewable once compromised. In addition, a compromised template can reveal the original biometric data, which constitutes a clear threat, as it can be used to track the user from one application to another. In fact, new algorithms have been proposed in the literature to reconstitute the original biometric trait by using the extracted features, attacking the stored features makes it possible to reconstruct the user’s fingerprint, face, or other biometric trait in order to usurp the identity of another person or entity. For these reasons, cancelable biometric template protection methods have been proposed to overcome these problems. Their basic idea is to transform the biometric data and achieve the matching in the transformed domain. The transform function must simultaneously fulfill the following four properties: non-invertibility, cancelability, accuracy and diversity. The non-invertibility property guarantees that the origi nal biometric data can’t, or it is hard to, be recovered even if some parameters of the transform function are known. This is generally ensured by the non-existence of the inverse transform function (one-to-many inverse-transformation) or by, simply, making the search-space size very large to escape to a brute-force attack. This thesis aims to investigate the robustness of cancelable biometric systems by analyzing and classifying various attacks targeting these systems based on well-defined criteria to enhance their security level. Additionally, it proposes a comprehensive evaluation framework grounded in stringent standards to assess the effectiveness of protection schemes against such attacks. Furthermore, a protection system for palmprint templates is implemented using irreversible transformation, ensuring a high level of security while ensuring the practical characteristics of these systems. On the other hand, we are interested in transformation-based techniques that establish a map ping between the original biometric template points and the transformed template. An attack against a cancelable fingerprint scheme is conducted to demonstrate the possibility of inverting the transformation function and to analyze the impact of correlation between multiple instances of protected templates generated from the same biometric trait on the efficiency of such algo rithms.