Author:

Abstract:

This thesis focuses on the analysis of techniques for the protection of biometric data. The increasing use of biometrics for the identification and authentication of individuals has raised many concerns because of the large-scale collection and automated processing of biometric data associated with it. These data can be abused, e.g., to gain unauthorized access to certain systems or to violate someone's privacy, and must therefore be protected. The techniques that have been proposed use widely varying approaches that often depend on the type of the underlying biometric characteristic, e.g., a fingerprint or the image of a face. This makes it a challenging task to evaluate their effectiveness.We analyze the irreversibility and unlinkability of techniques that are based on noise-tolerant transformations. Irreversibility is the property that a piece of biometric data, called a template, is hidden by transforming it in a way that cannot easily be undone. Unlinkability implies that it is infeasible to derive from two protected templates whether they originate from the same charateristic, e.g., two prints from the same finger. To evaluate these properties we develop different strategies based on, amongst others, heuristic inference and regression. Several successful attacks are demonstrated and some fundamental limitations are derived.More recent approaches rely on secure hardware elements or distributed protocols that use cryptographic primitives. To analyze these we use a blackbox attack model and we exploit the information that is revealed through the interaction with these blackboxes. We present a framework with generic attacks to facilitate the systematic evaluation of biometric authentication protocols in the malicious adversary model. This model is more realistic than the honest-but-curious adversary model in which most existing protocols have been designed.The diversity in approaches and the absence of universal metrics make it difficult to compare different techniques. Therefore, we analyze the criteria that are relevant for the holistic evaluation of different methods. With this consolidation we take a first step towards the independent benchmarking and ranking of biometric template protection methods.