Font Size: a A A

Study On Human Identification Algorithm Based On ECG

Posted on:2013-03-09Degree:MasterType:Thesis
Country:ChinaCandidate:L YangFull Text:PDF
GTID:2248330371961835Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
With the development of society economic and information, the traditional humanidentification technology couldn’t satisfied human’s growing require for security. The coming ofBiometric Identification Technology (BIT) effectively solve this problem, and BIT is applied inplenty of security fields, such as criminal investigations, financial transactions, electronic commerce,airport check-in, homeland security and so on. BIT utilized physiological or behavioralcharacteristics extracted from human subjects to complete personal identification. Fingerprint, face,iris and speech are biometrics characteristic commonly used. Compared to the traditional humanidentification technology, the BIT has lots of advantages. However, any BIT has its own drawbacks,it is easy to mimic by fake finger, iris and face photos etc. So, exploring a new biometric technologyto improve the system’s security becomes more important.This thesis studies on human identification algorithm based on electrocardiogram (ECG).Firstly, a pre-process algorithm was proposed to eliminate the effect of the noises and heart ratevariability (HRV), and measure the quality of ECG signals. And on this basis, considering fromtime domain and frequency domain some related ECG feature extraction algorithms were proposedto extract the discriminative characteristics from ECG signal. Finally, constructed patternrecognition algorithms, and some public ECG database like MIT-BIH ST change, PTB, QT wereutilized to evaluate these approaches, and high identification accuracy was obtained.The following works is accomplished in this paper:1、A pre-process algorithm was proposed. The mechanism and physiology of ECG was studied,and the noises’characteristics during recording process were analyzed. And then a ECG denoisemethod based on smoothness priors approach and wavelet minimax thresholding method and aECG normalized approach was utilized to eliminate the effect of low-frequency noises,high-frequency noises and HRV respectively. Next a quality measure algorithm was proposed basedon the ECG’s characteristic of quasi-periodic and periodicity transforms (PT). Finally, thepre-process method was verified through simulation on MIT-BIH database.2、On the basis of ECG’s temporal and spectral feature, a feature extraction algorithm based onEnsemble Empirical Mode Decomposition (EEMD) was proposed. Systematically learned the basicprinciples of Empirical Mode Decomposition (EMD) and EEMD, and analyzed the decompositioncharacteristics of EEMD. After the ECG signal was decomposed by the EEMD algorithm, thetemporal characteristics of key intrinsic mode functions (KIMFs) combined with their Welch powerspectrum were extracted by analyzing the spectrum distribution of ECG signal. These two kinds of characteristics were fused after the dimensionality reduction method principal component analysis(PCA) was done on them, and then these characteristic coefficients were putted into the nearestneighbor classification. Meanwhile, the affect of different select of coefficients like the number ofthe ensemble, noise variance, points of Fast Fourier Transform (FFT) and types of window to thealgorithm performance were discussed. Finally, this algorithm was evaluated through simulation onMIT-BIH database, and high recognition rate was achieved.3、Based on the time-frequency analysis of ECG signal, a FFT based matching pursuit(FFT-MP) algorithm for ECG feature extraction approach was proposed. Firstly, systematicallylearned the basic principles of matching pursuit (MP) algorithm, and then utilized FFT to improvethe MP sparse decomposition algorithm based on analyzing the structural property of theover-complete atom dictionary and the characteristics of ECG signal. Through analyzing thetime-frequency distribution of the atoms after sparse decomposition, we could find out that thetime-frequency distribution of the first few atoms could represent the main time-frequencycharacteristics of ECG signal. So the first three atoms’time-frequency coefficients and theirprojection were extracted as feature coefficients. And then, constructed support vector machine(SVM) to complete the identification. Finally, this algorithm was evaluated through simulation onMIT-BIH database, and high recognition rate was obtained.The research of this thesis provides a novel way for BIT with high accuracy, strong robustnessand good capability to prevent counterfeiting, which laid a solid theoretical foundation andtechnical support for the development of ECG identification system.
Keywords/Search Tags:Biometrics, Electrocardiogram, Feature Extraction, EEMD, Matching Pursuit, Classification
PDF Full Text Request
Related items