الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 140 |  |  |
| | | from | 140 | | |
Abstract
Recent reported research proved that Electrocardiogram (ECG) and Phonocardiogram (PCG) can be used as a biometric. In this work, we built and developed two different biometric identification systems one for (ECG) and the other for (PCG).
In the first system, we presented an ECG divided mainly into four stages namely, data acquisition, preprocessing, feature extraction and classification. First stage, data acquisition stage, data sets were collected from two different databases, ECG-ID and MIT-BIH Arrhythmia database. Second stage noise reduction of ECG signals using wavelet transform and series of filters used for de-noising. Third stage obtains the features using three different techniques a non-fiducial, fiducial and a fusion approach between them. In the last stage, the classification stage, three classifiers have been developed to classify subjects. The first classifier is based on Artificial Neural Network (ANN), the second classifier is based on Euclidean distance (ED) and the last classifier is sequential minimal optimization (SMO) algorithm for training a support vector machine (SVM) using polynomial kernel classifier. Classification accuracy of 95% for ANN, 99 % for ED and 99% for SVM on the ECG-ID database, while 100% for ANN, ED, SVM on MIT-BIH database.
In the second system we presented a machine learning approach based on feature level fusion for person identification using phonocardiogram (PCG). The proposed approach consists of five stages; starting with data acquisition, pre-processing, segmentation, feature extraction, and classification. Firstly data set were collected from the HSCT-11 database working on 60 subjects. Secondly process is concerned with noise reduction by removing noise from PCG signal using wavelets. Thirdly, segmentation process was done by applying Shannon energy envelope on the filtered signal to detect the positions of S1 and S2. Then a new combination of features is investigated for robust biometric PCG identification. Mel Frequency Cepstral Coefficients (MFCC) is efficient for PCG identification in clean speech while Wavelet features are robust for noisy environments. Therefore, combining both features together is better than taking each one individually the fusion is done using canonical correlation analysis (CCA). Finally, artificial neural network (ANN) and sequential minimal optimization (SMO) algorithm for training a support vector machine (SVM) using polynomial kernel classifier has been applied to classify subjects. The result shows a classification ratio 98.33%.
The results obtained from those two systems (ECG) and (PCG) were encouraging to show how robust our machine learning techniques used are. A comparison is already made with the previous work to ensure the efficiency of our methods.