Protecting the ECG Signal in Cloud-based User Identification System - A Dissimilarity Representation Approach

摘要

Biometric recognition has become a popular approach for user identification and authentication. However, since in ECG-based biometrics users cannot change their authentication/identification signal (unlike in password-based methods), its applicability is seriously constrained for cloud-based systems: a hacker could potentially retrieve the stored ECG signal, eternally disabling ECG-based biometrics for the attacked user. To overcome such an issue, new methodologies must be devised to enable cloud-based authentication/identification systems without requiring the transmission and storage of the user's ECG signal on remote servers. In this paper we propose an ECG biometric approach that relies on non-linear irreversible dissimilarity spaces to encode (encrypt) the user's ECG. We show how to construct the dissimilarity space, and also evaluate the system's accuracy with the dimensionality of the dissimilarity space. We show that the proposed biometric system retains similar identification errors as an equivalent system relying on the Euclidean space, while the latter can potentially be broken by using triangulation techniques to uncover the users original ECG signal.