Abstract: ECG-based based human recognition is increasingly becoming a popular modality for
biometric authentication. Two important features of ECG signals are liveliness and the robustness
against falsification. However, ECG features vary due to muscle flexure, baseline wander, and other sources of noise. This paper presents a new method which extracts multiscale geometric features from ECG signals and apply them for human identification. A non-linear filter is applied for preprocessing the ECG signal. The refined ECG signal is then divided into multiple segments and feature matrix is computed by multiscale pattern extraction technique. Feature matrix is finally applied to a simple minimum distance to mean classifier adopting leave-one-out procedure. An experiment with 60 ECG signals from 60 subjects shows promising performance of the proposed method compared to the conventional ECG features.
Keywords: Binary patterns, multiscale representation, supervised classification, and human
identification