Research On Mobile Phone User Authentication Based On Biometric Behavior Characteristics And Single Classification Algorithm

As an important tool for people's daily work and social communication,smartphones store a lot of information that is closely related to the privacy of users.Therefore,the loss or theft of a mobile phone can lead to a series of serious consequences.In this case,mobile phone user identity authentication has become an important security protection means.At present,the mainstream security verification measures in the smart phone security are password verification,such as personal authentication by PIN code,digital password or graphic password.However,these traditional verification methods are too simple and can easily lead to problems such as lost passwords or being easily broken.Although the method of using fingerprints or human faces to authenticate has improved the performance,it is easy for an attacker to use fingerprints,images,videos and other tools to crack.Moreover,the above authentication methods are all one-time authentication,that is,authentication is performed when the user unlocks the mobile phone.Users need an efficient,continuous authentication method to protect their phones for long-term authentication.This paper focuses on the issue of continuous authentication of smartphones.Based on the biological behavior habits of users when using mobile phones,the mobile phone's own behavioral information collection devices are used to construct biological behavior characteristics,and an implicit and continuous mobile phone authentication system is created.Most mobile phone authentication methods based on bio-behavior characteristics use multi-classification algorithms such as support vector machines.These algorithms have certain requirements for the capacity and balance of positive and negative data sets.In this paper,the single classification algorithm is used for the negative sample data set,and a mobile phone authentication system based on user biological behavior and single classification algorithm is proposed.The main research work includes:1.A continuous authentication system based on user touch behavior characteristics and Import Vector Domain Description(IVDD)single classification algorithm.Current smartphones rely on the user to touch the screen to interact,and the touch behavior of different users on the screen of the mobile phone is specific.The authentication method first collects information on the behavior of the user performing a touch slide on the screen,and one touch behavior includes a touch start point,an end point,a slide distance,and a force of a finger pressing the screen.For these touch information,32 touch behavior features were designed and extracted,and five features that are most helpful for user classification were selected and used for authentication.This paper chooses to train the new IVDD single classification algorithm to get the classifier of the legitimate user.The classifier can calculate a probability value for the touch behavior feature collected by the mobile phone.By comparing the set authentication threshold with the probability value,the authentication system can distinguish whether the user of the touch behavior is legal or illegal.In this paper,the authentication method is evaluated by the number of touch behavior features,computational complexity,and individual authentication accuracy,and compared with the traditional Support Vector Domain Desciption(SVDD)single classification algorithm.The experimental results show that when 15 touch behavior features are used for authentication,the authentication system can achieve an average error rate of 2.5%,and the error rejection rate reaches 2.14%.2.Based on mobile phone sensor and integrated single classification authentication system.The user's posture,direction and gestures when using the mobile phone have their own habits.The accelerometer,magnetometer and direction sensor that the smartphone comes with can receive these subtle behaviors.The method uses the user behavior information acquired by the three-axis sensor,extracts 48 behavior characteristics for each sensor data,extracts these features,and uses the SVDD single classification algorithm for authentication.Two authentication methods are proposed for authentication.The first is to fuse the characteristics of the three sensors,and comprehensively train a single classifier for authentication.The other is to train a single classifier based on three sensors.The results of the three single classifiers are fused.In this paper,the sensor datasets of 50 users are used for experimental verification.The results show that the authentication method for combining sensor features and the authentication method for the fusion of three single classifiers are better than the single authentication.