Study Of Human Identification By Electrocardiography Frequency Features

Counterfeiting ability of single biometric characteristic such as fingerprints, face and voice is improved, as well as the development of science technology, the authentication of personal identity attract widespread attention. So, search on a new biometric characteristic for identity authentication or combine kinds of biometric methods together is the direction of future biometric identity authentication. Heart is not only one of the most important organs of human, but also the one of largest intensity of electrical activity. Detection and analysis of electrical activity in the heart has become one of the most important method and means in the medical clinical practice detection and diagnosis of cardiac function one of the most important method and means. Because the ECG signals have advantages such as: universality, diversity, simple process, small volume, convenient data acquisition and stability of the longer-term and so on. In recent years, the identity authentication using human ECG has been a hot research area.In terms of research methods,At present there is two kinds of ECG signal identification algorithm: first, reference points detection based on ECG waveform feature extraction method, such as R wave amplitude、T wave amplitude、QRS intervals, QT intervals and so on. The time domain waveform features are easily affected by the change of heart rate and recognition rate is largely dependent on the accuracy of automatic identification of reference point in the time domain. Second, feature extraction method that is not based on reference points, such as using the method of ECG in the frequency domain.Methods of ECG frequency domain analysis have wavelet transform, discrete cosine transform and Fourier transform and so on. Because of the Fourier transform can well depict the amplitude frequency characteristics and phase frequency characteristics of ECG signal, and computing speed is fast, so in this thesis, the Fourier transform of ECG signal has been researched. Because of the technology of identity recognition based on ECG signal is a new technology, therefore there are many work to do to improve it. In order to improve the accuracy of identification and the efficiency of the algorithm, on the basis of summarizing the predecessors’ research results, this thesis mainly studied the method of ECG signal identification which is based on Fourier transform, feature extraction and screening the characteristics.Using correlation analysis and neural network classifier, confirming the feasibility of two methods by use of MATLAB simulation software. The main content is as follows:1、ECG signal data: This thesis takes the data of 10 people in the PTB database and 5 people in the MIT database, and to do the R wave detection for each data. 100 RR intervals are extracted for each data and Fourier transform is used on these data to get frequency domain features. So amplitude spectrum, phase spectrum and energy spectrum are obtained. The features contained frequency-domain signal slope, harmonic number, the magnitude gap, ratio of different frequency energy to total energy. Observing the differences between them.2 、 To extract the features: The features contained frequency-domain signal slope, harmonic number, the magnitude gap, ratio of different frequency energy to total energy and so on. Features are used to the principal component analysis and the contribution rate of characteristics is obtained.3、ECG signal identification algorithm: Classifiers of correlation analysis and neural network were used in identification.4、Improving the recognition rate: compared the result about two methods, summarizes the full text, put forward to the further research work, in order to ensure the stability of the identity, improve the rate of recognition.As we all know, correlation analysis is frequently used by identity recognition based on ECG, it is difficult to determine the threshold. To overcome the defect, this thesis puts forward to a new experimental method, comparing the correlation coefficient of test group and template correlation, the nearest number is regarded as the people in the template and effectively avoids the problem. Meanwhile, neural network were used in identification, compared with the correlation analysis method of the method, neural network achieves a good effect, but improving accuracy is still possible.