Research On Continuous Authentication Based On User Interaction Behavior Feature

For most existing computer systems,the internal resources of the system are available to the user when the user successfully login.After the user login the system,the system will not authenticate the user.At this time,the attacker may fake a real user and enter the system,causing a security crisis inside the system.Therefore,research on the continuous authentication(CA)of user identity is of great practical significance.In identity authentication,user biometric-based authentication has higher security and reliability.The identity authentication of the keystroke dynamics or mouse dynamics do not require additional equipment compared to other biometric authentication methods.In a large number of existing studies,CA based on keystroke dynamics or mouse dynamics are performed in a specific environment or requires the user to perform a specified action.But in real life,users will use the system in different locations,at different times and in different environments.The behavior of the user will vary with the context,and these differences may result in authenticating the real user as an imposter.In this thesis,exploring the possibility to perform CA without restriction,proposing a CA method based on user interaction information when context changes.The method adaptively provides different levels of security when the user login the system with different computers at different times and in different locations.At the same time,in order to prevent the attacker from avoiding detection by limiting to an input device,this thesis considers two behavior modes of the user using the mouse and the keyboard.In this method,firstly,collecting 30 users' data in an uncontrolled environment,secondly,quantifying user's context information,extracting the user behavior feature,and using fusion technology to identify users,finally,according to the recognition result and the quantitative context information judge whether the current user is a real user or not.The experiment result shows that the CA method can effectively implement the CA problem in different contexts.