ECG Biometric Recognition: Template-Free Approaches Based on Deep Learning.

Biometric technologies offer much convenience over the conventional approaches to identity recognition, but security and privacy concerns also accompany their applications. In this paper, an electrocardiogram (ECG)-based identification scheme is proposed to relieve such concerns. With the help of a deep learning (DL) technique, the identity of an unknown beat bundle can be determined without the need for biometric template construction. Thus, the disclosure of the physiological and pathological condition of an individual from his/her stolen templates will no longer be possible. Furthermore, the problem of being vulnerable to unregistered subjects in this DL-based recognition system is also addressed. Experiments with real and synthesized ECGs are used to illustrate the efficacy of the proposed scheme. An identification rate of 97.84% for the 200 registered subjects with a false-positive identification rate of 0.69% under the attack of 1,000 synthesized single-lead ECGs was achieved.