Identity Authentication On Smartphone Based On The User's Behavior

With the popularity of the intelligent applications on smartphones,privacy protection is a non-negligible and critical issue,which is mainly achieved by identity authentication.The existing solutions to this problem are mainly using a password to lock the screen or utilizing the distinct unlocking gesture features.However,password-based methods are inherently vulnerable to shoulder surfing attacks and it cannot distinguish the owner and attackers who access the smartphone from stealing the password.Besides,both the owner and the borrower are likely to use the phone after the owner finishes unlocking the phone,so the person who unlocks the smartphone is not definitely the person who uses the smartphone,yet gesture-based schemes cannot distinguish the owner and the borrower.Also,both the solutions just identify whether the user is the owner or not.However,based on such authentication result,it cannot prevent the borrower from viewing the owner's privacy,so the owner's privacy protection requirement cannot be satisfied.Under the background of privacy protection,we propose a identity authentication approach for three kinds of users including the owner,the borrowers and the attackers according to the user's behavior,which focuses on addressing the following challenges:1)The behaviors are quite similar among different users,which makes it difficult to distinguish.2)The behavior lasts for very short time,and there exists few information for reference in identity authentication.3)Making the identity authentication approach be robust to the diversity and randomness of the usage scenarios is another big challenge.Unlike past work,which only utilizes unlocking behavior for identity authentication,our design takes both the unlocking behavior and the motion sequence into consideration.We decompose user's behavior into several independent motions and use the motion sequence that consists of different motions to express the usage habit of different user.First,based on Markov Chain,we design an algorithm to estimate how likely the motion sequence is performed by the owner.Then through effective feature selection,we determine how likely the unlocking behavior is performed by the owner via logistic regression analysis.Finally,based on the difference in motion sequence and unlocking behavior,and the context information in the motion sequence,we propose identity authentication algorithm and achieve identity authentication for three kinds of users.We evaluate the performance in real environments.The experimental results show that the accuracy is more than 94%with robustness in various usage scenarios,which is a key step towards privacy protection.The identity authentication model we proposed can be extended in other safety fields based on behavior difference,e.g.,anti-theft of the smartphone.