Research On Keyboard Guessing Attack And Defense Technology Based On Mobile Phone Sensor

With the development of science and technology,mobile phones are no longer just used for talking,but need more powerful functions.In order to make mobile phones more intelligent and humanized,more and more sensors are gradually embedded into smart phones.Most of the calls of these sensors do not require users to apply for permission,which greatly facilitates the work of developers,but on the other hand,it also causes the leakage of privacy,which gradually attracts more and more people’s attention.Keyboard input is a high-frequency operation of mobile phone users,which usually contains important user privacy information such as password.At present,there have been many researches on keyboard guessing attacks based on mobile phone sensors,but in the existing keyboard guessing attack methods based on mobile phone sensors,the classification method is only a simple base classifier.In view of this situation,this thesis improved the existing keyboard guessing attack method based on mobile phone sensors.Through the in-depth study of attack technology,two kinds of security strategies are proposed to reduce the threat of this kind of side-channel attack.Finally,in view of the problem that the sensor data collection is too complicated,the generative adversarial networks is used to expand the sample data set.The main work contents of this thesis are as follows:(1)On the previous researchers’ method of keyboard guessing attacks based on mobile phone sensors,this thesis has improved the classification and put forward a method of ensemble learning based on the algorithm of decision tree,naive bayes,support vector machine(SVM)and K-nearest neighbor.This method gives better classification results.In addition,a click detection module is added to screen sensitive data,and a user click detection method based on energy threshold is proposed.(2)On the research of keyboard input guessing based on mobile phone sensors,this thesis studies how to defend against this kind of side channel attack,and proposes a security mechanism based on Black List & White List.In addition,a security policy based on noise interference to reduce the threat of this kind of side channel attack to user privacy is proposed.(3)The method of generative adversarial networks is proposed to expand the collected sensor time series data set,so that the small data set can be expanded into a larger data set and the data set used in the training model is more abundant.To some extent,the tedious data collection of mobile phone sensor data can be avoided.