Research On The Identity Authentication Technology Based On Keystroke Dynamics

User authentication technology is the main way to ensure information security.Conventional password authentication methods can only verify information like password,and may suffer from security defects like password leakage.Physical biometric identification technologies such as fingerprints and face recognition require the integration of special equipment,and the abuse of physiological characteristics information has caused people to worry about privacy issues.Keystroke dynamics,as a behavioral feature,is difficult for others to imitate,and the keystroke behavior data can be continuously obtained without additional equipment.Therefore,this thesis studies keystroke dynamics from two aspects: fixed text and free text.In the research of user authentication technology based on the keystroke characteristics of fixed text,since the keystroke behavior is easily affected by physical,mental and environmental factors,this paper finds that the keystroke time sequence of fixed text input is non-stationary through unit root test.To this end,the keystroke sequence is transformed into a stationary sequence through the second-order difference.In view of the fact that wavelet transform can automatically adapt to the requirements of signal time-frequency analysis and highlight the local information of the signal,this thesis does a multi-scale discrete wavelet analysis on the sequence of difference,and finds that the high-frequency features are quite different among users,and the low-frequency features are stable in distribution.Because multi-level decomposition is a re-decomposition of low-frequency signals,the approximate coefficients and detail coefficients obtained by the single-level discrete wavelet transform are used as keystroke features.Based on Manhattan distance,this work proposes a two-factor authentication method combining keystroke dynamics and password.Experimental results show that the equal error rate of this work is 3.1%,which is a certain improvement over recent researches.The average time of computation for a single authentication is 12.2milliseconds,which satisfies the need of fast authentication.In the research of user authentication technology based on the keystroke characteristics of free text,related studies have shown that the high-frequency key pair features in free text are stable,and thus can be used to effectively distinguish users.According to statistics,the10 most frequent key pairs for all users in the data set of this thesis are: ‘AN'?‘NG'?‘JI'?‘EN'?‘SH'?‘IA'?‘IN'?‘XI'?‘ZH' and ‘Backspace-Backspace'.Analysis found that the characteristics of these key pairs differ significantly among users.Due to the large amount of free text data and complex features,this paper takes the time feature of the abovementioned key pairs as the free text keystroke feature in order to reduce the complexity of the system.And isolated forest model is trained for each group of features.Considering that each key pair has a different effect on distinguishing users,this paper uses the accuracy of each classifier as the weight to design a continuous authentication system based on the system trust score mechanism.The experimental results show that the false acceptance rate of this model is 3.4%,and the false rejection rate is 2.5%,which is a great improvement compared with recent researches.Both the authentication system based on fixed text keystroke characteristics and the user authentication system based on free keystroke characteristics designed in this paper have higher accuracy.The former performs auxiliary authentication when the user logs in,and the latter continuously monitors the user's biometric identity after login.The combination of the two measures help enhance information security by improving the accuracy of authentication.The keystroke sequence frequency analysis method proposed in the study of fixed text provides a new direction for the study of keystroke dynamics.