Abstract

The development of biometric system using Electromyogram signals is discussed in this paper. The EMG signals\nare the physiological signals generated due to the neuromuscular activity. The EMG signals reflects the strength of muscles\nof the person, which may vary from person to person. The Non Uniform Filter Bank cepstral features are proposed for the\nrepresentation of person specific information present in EMG signal. The variation of EMG signals from person to person is\nmeasured by using Euclidian distance metric for the NUFB cepstral features and KL Divergence distance metrics for GMM\nmodels of NUFB features. The average KL Divergence based inter person variability is 77.19%, whereas intra person\nvariability is 44.19%. The GMM-EM and GMM-UBM systems are developed for 50 persons EMG data using NUFB\nfeatures. The performances of both systems are analyzed by varying the number of training time slots per person and number\nof Gaussian Mixtures. The GMM-UBM gives an accuracy of 94.12% for two EMG time slots of 10 seconds duration,\nwhereas GMM-EM requires 6 time slots per person to give the same performance.