Continuous Authentication For Mobile Users Based On Feature Fusion

Smart phones have gradually become a necessity in daily life.At the same time,the privacy and security issues involved have been paid more and more attention.Identity authentication technology is one of the important means to protect the privacy and security of mobile phone users.At present,the identity authentication methods on smart phones mainly include password based authentication represented by digital password and physiological biometric authentication represented by fingerprint or face.However,these authentication methods belong to one-time authentication,which cannot verify the user's identity in the process of using the mobile phone.The authentication based on behavioral biometrics,which is implicit and persistent,is based on the user's behavioral information automatically obtained by the built-in sensor in the process of using the mobile phone.However,it is often limited by poor data quality,noise and large internal differences to only authenticate a single behavioral biometric.Using of feature fusion technology to fuse multiple modal features can realize the complementary advantages of each feature,and then make the authentication with higher accuracy and stronger adaptability.This paper studies the continuous authentication based on behavior features,and proposes two continuous authentication schemes based on feature fusion.The main research work of this paper includes:1.Continuous authentication based on balanced feature association.A convolution neural network is designed to solve the problems of time-consuming,laborious and limited performance in artificial feature design.The depth feature is extracted from the time domain and frequency domain data of accelerometer and gyroscope by using the network.When fusing the extracted features in time domain and frequency domain,considering the shortcomings of traditional serial and parallel fusion strategies,a balanced feature connection strategy is proposed to perform fusion,so as to obtain more expressive fusion features.Finally,based on the fusion features,the one-class SVM classifier is used for user classification authentication.2.Authentication based on adaptive weight feature fusion.The feature fusion process is integrated into convolutional neural network,and a convolutional neural network with three network streams is designed.Each network flow in the network extracts features from the time domain data of accelerometer,gyroscope and magnetometer respectively,and then the feature fusion with adaptive weight is carried out in the feature fusion layer.Finally,after feature selection,the fusioned features are authenticated by using isolation forest classifier.3.The two authentication systems are evaluated by experiments.The results show that the equal error rates of the two authentication systems are 1.0% and 1.5%respectively,and the authentication time are both about 5s.Compared with other authentication systems,they both have better time efficiency and authentication accuracy.