Efficient privacy-preserving biometric identification in largescale multibiometric systems

Abstract

In recent years, applications of biometric systems on national and international scale have appeared. Biometric identification is one of the important operational modes of such systems. It entails ascertaining the data subject identity corresponding to a given biometric sample, solely using the information from said biometric sample, i.e. effectively conducting a nearestneighbour search. The na¨ıve search method, i.e. an exhaustive (linear) search of the biometric enrolment database, suffers from two drawbacks, namely: high computational workload and increased probability of false positive occurrences. Consequently, research into computationally efficient methods of biometricidentification is necessary; it is the main topic covered in this thesis. Specifically, the key contributions of this thesis are: • Formulation of a taxonomy for conceptual categorisation of methods of efficient biometric identification. A comprehensive survey of the relevant existing publications and organisation thereof in the context of the developed taxonomy. • Development of methods which substantially decrease (by space search and/or template comparison cost reduction) the computational workload requirements of the biometric identification transactions, including: – Methods which take advantage of the intrinsic properties of certain types of biometric characteristics and/or biometric feature representations. – Methods which can be applied irrespective of the type of biometric characteristic and the biometric feature representation. – Methods which utilise biometric information fusion. • Development of methods (both general purpose and biometric characteristic specific) of biometric template protection in the aforementioned context of computationally efficient biometric identification systems. • Development of methods relevant to other (e.g. stress testing and usability) aspects of the operational biometric identification systems.