Study Of Electrocardiogram Key Features For Human Identification

Since modern society has over relied on Internet, human identification has become an important safety issue in this virtual electronic society. Different from the traditional password, made up by digit and symbol, biometric( like fingerprint, iris, face, voice, gait, palm print, etc...) based identification have been widely applied at present. But, these biometric can be copied in some ways. Electrocardiogram(ECG) is the biometric with "Living", which has gradually become a hotspot in the research area.The key study content of ECG feature based identification is feature selection and feature recognition, they directly influence identification accuracy rate. In the paper, ECG features computed by ECG waveform fiducial points are used for identification. The paper proposed a key feature set extraction strategy for ECG identification. The strategy contained three parts, initial feature set selection, feature ranking by contribution rate and key feature set selection.After literature review working on the common achievements in this field and standard description of ECG feature in clinical medicine, four types with 26 features are determined as initial feature set, it covers interval feature( like heart rate, QRS interval, etc...), amplitude feature( like the voltage of P wave, etc...), slope feature( like slope of Q and R, etc...) and area feature( like area of triangle SS’T). In the study, ECG waveform fiducial points were signed manually, and features were computed by these fiducial, furthermore, stepwise discriminant analysis was used to conclude feature ideal contribution rate in ECG identification.In the next step, the paper used wavelet transform to detect ECG waveform fiducial point automatically, and computed the detection rate. This is to find out the influence of fiducial point automatic detection on ECG identification. After that, the detection rates were used to compute the confidence rate(accuracy rate) of features. Furthermore, combined with the ideal contribution rate, real contribution rate of features were finally concluded, and so did the stable key feature set.In the paper, following experiment data set were used: 1, In PTB(Physikalisch-Technische Bundesanstalt, an ECG database provided by German National Institute of Metrology for all kinds of medical diagnosis)database, we selected 60 healthy testers’ 60 seconds V5-lead ECG data; 2, 120 healthy undergraduate volunteers’ 90 seconds V5-lead ECG data were collected by self-designed synchronous12 lead ECG device. By the proposed strategy, the amount of key features set was 9 and 17 of each ECG dataset respectively. The identification accuracy rate reached 99.7% and 94.8% respectively.