Research Of Behavior-based User Authentication On Mobile Device

Recently,smart mobile devices such as smartphones,smartwatchs and smartglasses have been widely involved in our daily life.A large amount of private and sensitive data are stored in the device.In order to enhance the security of mobile device,a secure and convenient user authentication scheme is desired.However,the existing authentication schemes have some limitations on commercial mobile devices.For example,the password-based and pattern-based authenticaiton may be attacked by smudges and shoulder-surfing.The voiceprint-based and face-based authentication can be spoofed by replay attacks.Rapid advances in semiconductor fabrication technology have enabled the abundant sensors to be equipped to the mobile devices.The sensor records a wealth of behavioral data while the user operates the devices.The human behaviors have a huge connection with lifestyle and muscle structure.The behavior-based user authentication approach has receieved great attentation in both industry and academia.In this thesis,we propose two behavior-based user authentication schemes based on the user behavior characteristics of the smartphones and the smartwatchs.The main research work consists of two parts:1)Implicit Authentication for QR Code Payment behavior using Inertial Measurement Unit on smartphone.It is the first user authentication scheme designed for the behavior of mobile payment.The data capture by the accelerometer and gyroscope are used to authenticate users when they perform the QR code payment.Firstly,we decompose the motion into two sub-actions:"pick up"and "put down" by the processing of behavior.The behavior features are extracted from the two sub-actions respectively.In addition,we analyze the discriminability of features,and use the combined feature dimension reduction algorithm to remove some low-quality features.Finally,some machine learning algorithms are leveraged to build classifiers to verify the feasibility of scheme.Compared with the existing authentication scheme based on the sensors of smartphone,the experimental results show that our shceme can achieve an average authentication accuracy of 94.86%,and authenticate the user's identity in a shorter time.2)User Authentication from smartwatch Photoplethysmography(PPG)sensor using gesture behavior.The scheme can reduce the number of registration samples required by the user.The PPG sensors of smartwatch are used to collect the finger-level gesture behavior signal.The user's identity are authenticated explicitly when they perform the predefined gesture.We first propose a new method to detect the gesture segments from the raw PPG signals.Then,the Siamese Network-based authentication model is developed to determine the identity of user.In the siamese network,a special neural subnetwork based on the feature pyramid network(FPN)and long short-term memory(LSTM)network is designed to extract user gesture feature.Finally,we conduct some experiments to evaluate the performance of authentication model.The experimental results show that our scheme achieves the authentication accuracy above 90%with only one prior instance.In addition,the authentication model achieves an average authentication accuracy of 92.43%compared with the existing schemes.