Implementation of ECG Based Person Authentication System Using NUFB Features and GMM-UBM

The implementation of person Authentication system using Electrocardiogram (ECG) is discussed in this paper. The ECG signals are the physiological signals represents the electrical activity of heart. The ECG signal is unique to each individual and different for different persons and inter variability in different subjects. Experimental proved that physiological factors such as conductivity of various areas of heart, the heart mass orientation and other factors like lead placement. Timing of depolarization are unique from person to person. The Non Uniform Filter Bank cepstral features are proposed for the representation of person specific information present in ECG signal. The variation present in the ECG signals is measured by Euclidian distance measure for the NUFB cepstral features and the KL Divergence distance metrics used in the GMM models of NUFB features. The average KL Divergence based inter person variability is 79.59%, whereas intra person variability is 36.73%. The GMM-EM and GMM-UBM systems are developed for 49 Subjects ECG data using NUFB features. The performances of both systems are analyzed by varying the number of training time slots per person and number of Gaussian Mixtures. The GMM-UBM gives an accuracy of 87.07% for two slots of ECG data, whereas GMM performance is 77.55% which is the less than GMM-UBM Performance for same number of slot of data.