Study Of ECG Biometrics Based On Non-negative Matrix Factorization

As a vital sign,Electrocardiogram(ECG)has highly discriminative characteristics in the field of biometrics.Some studies show that ECG signals are different among different individuals,and ECG signal will not change greatly in a period of time.With the development of micro sensor technology,the collection of ECG signal is more convenient.Therefore,the ECG signal meets the premise of biometrics,is a relatively safe and reliable identification technology,and has good application value.Although the ECG biometrics has some advantages,the existing ECG biometric technology still has many problems to be overcome.To solve the above problems,this thesis proposes two biometric methods based on non-negative matrix factorization(NMF),one is based on non-negative matrix factorization,the other is based on graph regularized non-negative matrix factorization(GNMF)and sparse representation.In the ECG biometrics based on non-negative matrix factorization,the QRS wave of ECG signal is enhanced by the reconstruction of ECG signal,which is helpful to extract the discriminative features of ECG signal.The reconstructed ECG signal is transformed into a positive ECG signal by adding a positive value to satisfy the non-negative constraint of non-negative matrix factorization,which ensures that the elements in the basic matrix and coefficient matrix are both positive and the matrix is sparse.In the process of non-negative matrix factorization,not only the non-stationary signal can be factorized,but also the factorization level is not considered.In the process of calculation,the basic matrix and coefficient matrix can also be initialized randomly,so as to reduce the dimension of ECG signal,get the essential characteristics of ECG signal,and improve the effect of ECG biometrics.It can be seen from the experiment that the biometric method of ECG signal based on non-negative matrix factorization can effectively extract the characteristics of ECG signal,remove the redundant data of ECG signal,and have better recognition performanceGNMF and sparse representation methods implement ECG classification by performing graph non-negative matrix factorization and sparse representation on ECG signals.GNMF can encode the geometric information of ECG data and add label information to realize the dimension reduction and ECG primary feature extraction Sparse representation is utilized to obtain final features to perform ECG matching.This method can capture the important information of ECG biometrics,reduce the intra-class differences and increase the inter-class differences of ECG.The recognition rate is up to 100%on both MITDB dataset and ECG-ID dataset.As the number of fusion heartbeats in a test sample increases,the value of the EER almost drops to 0%on two datasets.Experiments show that this method can obtain robust ECG features.