Human identification by quantifying similarity and dissimilarity in electrocardiogram phase space

Specific patterns of electrocardiogram (ECG), along with other biometrics, have recently been used to recognize a person. Most ECG-based human identification methods rely on the reduced features derived from ECG characteristic points and supervised classification. However, detecting characteristic points is an arduous procedure, particularly at low signal-to-noise ratios. The supervised classifier requires retraining when a new person is included in the group. In the present study, we propose a novel unsupervised ECG-based identification method based on phase space reconstruction of one-lead or three-lead ECG, saving from picking up characteristic points. Identification is performed by inspecting similarity or dissimilarity measure between ECG phase space portraits. Our results in a 100-subject group showed that one-lead ECG reached identification rate at 93% accuracy and three-lead ECG acquired 99% accuracy.