A System of Biometric Authentication Based on ECG Signal Segmentation

One of the fundamental difficulties of ECG (Electrocardiogram) based biometric systems is intrabeat variation. In order to decrease these variations and increase the performance of the biometric system, we have proposed a new convolution based method for beat extraction and a waveshape based method for beat segmentation. In the feature extraction stage, thirty spatial and interval features are extracted and they are categorized in six groups using Feature Forward Selection (FFS) method. Feature classification is done by using four classifiers: Nearest Neighbor, Gaussian, Principal Component and Parzen Window data description. The experiment is done on MIT-BIH normal sinus rhythm database and the proposed method is achieved to %2.34 +/- 0.19 Equal Error Rate (EER) and %99.73 +/- 0.04 Area Under the ROC Curve (AUC) using Parzen Window classifier.