Ecg Signal Based Authentication And Implementation On Mobile Terminal

With the rapid development of technology,people have had high requirements for the security of information.Traditional identification methods such as digital password can hardly meet the high demand of today’s society.Under this circumstance,biometrics become widely used in the field of identity recognition for they are not easy to forget and lose.Compared with the common biometrics such as face,fingerprint and iris,ECG has the advantages of procurable,tractable,difficult replication and liveness detectable,and has unique application prospect.At present,a lot of progress has been made in the research of ECG authentication methods at home and abroad,but there are still some shortcomings,for example,the chronergy of algorithm is low and the scale of experimental data is small,so it is difficult to meet the actual needs.Besides,most of these studies are only focus on the algorithm,and do not realize a mobile terminal of authentication system.In view of these problems,the main contents of this paper are as follows.(1)The two authentication methods are based on Euclidean distance between beat features.The recognition task is changed from multi classification based on signal features to two classification based on similarity between signals.The homo and hetero recognition is carried out according to the distribution of Euclidean distance between features.The first is a method based on quantile of distance between features.When registration,registrants’ quantile of Euclidean distance is used to set threshold.When authentication,the data set of Euclidean distance from registrants and authenticators is calculated for matching.When the data set meets the threshold,it is determined that the authentication is passed.The second is a method based on probability density of distance between features and random forest.When registration,Gaussian kernel density estimation method is used to calculate the probability density of Euclidean distance between the features of the registrants,which is input into the random forest for training.When authentication,the probability density of Euclidean distance between the features of the registrant and the verifier is calculated and input into the random forest for testing.When training and testing,the label of the same person’s probability density is set to "1",and the label of different person’s probability density is set to "0".(2)The two methods are verified in A and B data sets respectively.Data set A is from a public database,including 1000 short-term ECG signals of 1000 persons.Data set B is from a self built database,including 261 short-term and long-term ECG signals of 87 persons.The best test results of the method based on the quantile of distance in data set A and data set B are 0.98 AUC and 0.0313 EER.It is found that AUC decreased to about 0.93 and EER increased to about 0.15 after testing in long-term ECG signals of data set B.The best test results of the method based on probability density of distance are 0.97 AUC and 0.10 EER in long-term ECG signals of data set B,and the accuracy is improved compared with the former method.(3)The algorithm is transplanted into a portable Jetson TX2 hardware platform,and the function of mobile terminal is verified.To sum up,the authentication methods designed in this paper use statistical information of similarity between ECG signals,which not only has high accuracy in large-scale short-term data,but also shows high performance in long-term data.They can be realized in mobile terminal.This study provides technical support for the research and development of ECG authentication system.