Protezione dei template biometrici per sistemi di autenticazione basati su firma

One of the most emerging technologies for automatic people recognition is biometrics. In contrast with traditional approaches, based on what a person knows (password) or what a person has (ID card, tokens), biometric based authentication relies on who a person is or what a person does. Biometric based recognition systems are then typically able to provide improved comfort and security for their users, when compared to traditional authentication methods. Unfortunately, the use of biometric data in an automatic recognition system also involves various risks not affecting other methods: if biometric data are somehow stolen or copied, they can be hardly replaced. Moreover, biometric data can contain relevant information regarding personality and health, which can be used in an unauthorized manner for malicious or undesired intents. It is also worth pointing out that, when a cross-matching among different biometric databases is performed, an unauthorized user tracking of the enrolled subjects can be done by means of users' biometric traits. This would unavoidably lead to users' privacy loss. Therefore, when designing a biometric based recognition system, the issues deriving from security and privacy concerns have to be carefully considered. Moreover, the adopted countermeasures should enhance biometric data resilience against attacks, while guaranteeing acceptable recognition performance. This Thesis is focused on the protection of the biometric templates employed in a sig- nature based authentication system. Signature biometrics is usually characterized by a high intra-user variability and a small forgeries inter-user variability, thus representing a challenging field of application for template protection techniques. The literature regarding biometric template protection and on-line signature based recognition is first reviewed. Then, we take into account both parametric and functional features based on-line signature verification approaches, and describe, for each of them, how to provide protection to the employed biometric templates. Specifically, we propose the use of cryptographic techniques and error correcting codes to secure global parametric features extracted from an on-line signature. Together with protection, also template cancelability and renewability are guaranteed. Moreover, the proposed authentication scheme is tailored to the signature variability of each user, thus obtaining a user adaptive system with enhanced performances with respect of a non-adaptive one. We then propose how to provide security to the templates employed in a functional feature based signature authentication system, by means of a feature transformation protection approach. Specifically, we introduce a set of non-invertible transforms, which can be applied to any sequence based biometric template to generate multiple transformed version of it. Retrieving the original data from the transformed one is computationally as hard as random guessing. The effectiveness of the proposed approach is tested by considering both a regional signature functions analysis (employing Hidden Markov Models) and a local signature functions analysis (employing Dynamic Time Warping). Moreover, the performances achievable with the fusion of these two approaches are also discussed. Eventually, we also propose the use of watermarking techniques to protect a set of dynamic signature features, by embedding it into a static representation of the signature itself. User authentication can be performed either by means of the only signature static image, or by using it together with the dynamic features embedded in the enrollment stage, by using a fusion approach. A multi-level authentication system, which is capable to provide two different levels of security, is then obtained. The proposed watermarking techniques are based on the properties of the Radon transform, being thus tailored to images, like those of a signature, with sharp edges. A procedure for the selection of the dynamic features which allow to guarantee the best recognition performances, as well as a novel approach which defines the minimum number of bits which should be employed to binarize a given feature without affecting the recognition performances, is proposed. The effectiveness of the proposed approaches is tested by employing the public MCYT on-line signature corpus, with signatures taken from 100 different subjects, as experimental database.