Research On Implicit Authentication Of Intelligent Mobile Devices Based On Behavior Characteristics

Smart phones and other mobile devices are widely used in people's daily activities.The leakage of personal accounts and other privacy information will cause huge losses to users.The purpose of authentication mechanism is to identify fake users and prevent them from unauthorized access to devices.Implicit authentication is the use of user behavior and habits for identity authentication.Because of its convenience,user friendliness and security,it has become a new research hotspot in the field of intelligent mobile device authentication.However,the existing researches usually use the methods of designing complex algorithms,extracting high-dimensional features,improving model complexity and combining with wearable devices to improve model accuracy and authentication performance.This not only makes the authentication mechanism cumbersome,but also increases the cost,and ignores the usability and practicability of the scheme.At the same time,mobile devices have inherent computing and processing limitations.In view of the above problems,the thesis explores the authentication principle based on behavior features,analyzes the causes of the differences of behavior features in detail,studies the distinguishing points between behavior features,and proposes three implicit authentication schemes based on behavior features from the research points of sensor data fusion,improvement of authentication algorithm and behavior differences.The main contents of the thesis are as follows:1.Implicit authentication based on multi-sensor behavior adaptive features.Based on the hand behavior characteristics,the perception differences of accelerometer,gyroscope and magnetometer for the same behavior characteristics are analyzed.According to the different hand actions perceived by sensors,the behavior characteristics corresponding to the sensor are selected to maximize the authentication performance of the selected features.In addition,the thesis combines the behavior characteristics of multiple sensors to study which combination of behavior characteristics has higher authentication performance.The experimental results show that the performance of accelerometer and magnetometer is better,compared with the related work,it has higher accuracy,and the scheme can be applied to a certain level of privacy protection.2.Multilevel implicit authentication mechanism based on dynamic trust value.Through the analysis of the historical results of model authentication,it is found that the authentication accuracy of real users has a certain regularity,which is usually stable in a range.Therefore,the thesis establishes a multi-level implicit authentication mechanism based on dynamic trust value combined with user history authentication results.On the basis of the authentication model trained by the behavior characteristics of stroke screen,the change trend of the user's historical authentication results is analyzed,and the mean value of the historical authentication results is taken as the dynamic trust value.The scheme sets the primary classification model as the former level authentication and the dynamic trust value as the latter level detection.Compared with the related work,the security of implicit authentication is enhanced.In addition,in order to keep the trust value reflecting the latest trend of user authentication in real time,an array of historical authentication records is established in the system.The historical records are updated by sliding window,and the mean value of historical results is calculated to update the trust value dynamically.3.Continuous authentication of mobile devices based on multi-behavior interaction.Implicit authentication itself is based on the law and uniqueness of behavior habits.Studying the differences of user's different behavior patterns helps to improve the performance of authentication model.The thesis analyzes two common behavior interaction modes among users: hand posture feature and hand micro action feature.According to the different interaction patterns,the authentication reliability of the two behavior patterns and the authentication performance of the multi behavior interaction patterns are analyzed.At the same time,the thesis conducts behavior imitation attack experiments to further evaluate and verify the performance of multi behavior interaction mode.The proposed implicit authentication scheme has the advantages of convenient data collection,easy feature extraction,no need of additional equipment and fast model training,and has high practicability,which can be applied to different security level scenarios.