Heart sound is generally used to determine the human heart condition. Recent reported research proved that cardiac auscultation technique which uses the characteristics of phonocardiogram (PCG) signal, can be used as biometric authentication system. An automatic method for person identification and verification from PCG using wavelet based feature set and Back Propagation Multilayer Perceptron Artificial Neural Network (BP-MLP-ANN) classifier is presented in this paper. The work proposes a time frequency domain novel feature set based on Daubechies wavelet with second level decomposition. Time-frequency domain information is obtained from wavelet transform which in turn is reflected in wavelet based feature set which carries important information for biometric identification. Database is collected from 10 volunteers (between 20-40 age groups) during three months period using a digital stethoscope manufactured by HDfono Doc. The proposed algorithm is tested on 4946 PCG samples of duration 20 seconds and yields 96.178% of identification accuracy and Equal Error Rate (EER) of 17.98%. The preprocessing before feature extraction involves selection of heart cycle, low pass filtering, extraction of heart cycle, aligning and segmentation of S1 and S2. The identification is performed over the score generated output from the ANN. The experimental result shows that the performance of the proposed method is better than the earlier reported technique, which used Linear Band Frequency Cepstral coefficient (LBFCC) feature set. Verification method is implemented based on the Mean square error (MSE) of the cumulative sum of normalized extracted feature set.
展开▼
