Research On Auxiliary User Authentication Of Smart Phones Based On Multi-modality Fusion

With the widespread use of smartphones and the continuous enrichment of functions,a large amount of personal privacy information is stored in smartphones.If private information is stolen or abused,it may pose a major threat to the reputation,property,and even personal safety of users.User authentication is an important method to protect personal privacy,but traditional authentication methods have certain limitations.For example,passwords are easy to forget or be peeked,patterns are easy to be snooped,fingerprints and face information are easy to be copied.In addition,commonly used authentication methods only realize static authentication when unlocking the phone.Due to the above-mentioned problems in traditional authentication methods,the academia and industry have begun to pay attention to continuous authentication technology based on user behavior.At present,the research on authentication technology based on user behavior is still in its infancy.The low authentication accuracy and robustness are the two major bottlenecks that restrict the practical application of this technology.Therefore,the paper studies the problem of continuous authentication based on user behavior characteristics,and focuses on the research of multi-modality-fusion-based auxiliary authentication technology for smart phones.In order to improve the accuracy of continuous authentication based on user behavior,this paper proposed an auxiliary authentication method based on multi-modality fusion(MMF).The proposed MMF auxiliary authentication method can continuously and transparently authenticate the current user after the user uses the primary authentication method to unlock the phone.The proposed method integrates the three types of feature data of the swiping screen modality,keystroke modality and smartphone motion modality when the user interacts with the phone to improve the authentication accuracy.The MMF method adopts the Isolation Forest algorithm to detect illegal users.In order to enhance the robustness of the proposed MMF auxiliary authentication method,the paper proposed a multi-modality fusion auxiliary authentication method based on posture awareness(MMF-PA).The proposed MMF-PA auxiliary authentication method adopts a two-level classification structure,which first uses two posture classifiers to recognize the current posture of the user,and then inputs the feature vector into the corresponding user classifier of the current posture to authenticate the user's identity.This two-level classification structure enhances the authentication method's robustness to the changes of user body posture,and further improves the accuracy of authentication.In order to verify the performance of the proposed multi-modality-fusion-based user auxiliary authentication methods,two customized Android applications were developed to collect user behavior data.We collected the single-finger swiping data,continuous keystroke data and built-in sensor data of 33 volunteers,and used the data from 25 volunteers to create a user data set.On this basis,the parameters of each module of the auxiliary authentication methods were determined,and the performance of the auxiliary authentication methods was tested and evaluated.The experimental results show that when the decision window is set to 9,in a free posture,the MMF auxiliary authentication method achieves an accuracy of 95.33%,and the MMF-PA auxiliary authentication method achieves an accuracy of 99.34%.The two multi-modality-fusion-based auxiliary authentication methods proposed in the paper have good authentication accuracy.In particular,the proposed MMF-PA auxiliary authentication method enhances the robustness of MMF-PA auxiliary authentication method to user posture changes,and has a good application prospect in the field of continuous user identity authentication of smart terminals.