Using Recurrence quantification analysis and Generalized Hurst Exponents of ECG for human authentication

Abstract

Previous works show that the electrocardiogram is a promising signal to be used as a biometric trait. The nonlinear methods for computing the dynamical properties of ECG signal, have been previously used. Since each of the large scale features of recurrence plots of ECG is related quite simply to time-domain features, they can provide good result in biometric system. In this paper we apply Rescaled Range Analysis (RSA), Higuchi's Fractal Dimension (HFD), Detrended Fluctuation Analysis (DFA), Generalized Hurst Exponent (GHE) and Recurrence quantification analysis (RQA) to extract features for authentication system. Support Vector Machine is used to classify the nonlinear features. The proposed approach has been tested using 18 different subjects ECG signal of MIT-BIH Normal Sinus Rhythm Database. The obtained results show that the authentication accuracy is 96.07±0.86%.