Research On Adaptive Website Fingerprinting Attack And Defense

Anonymous communication is a technology to protect the identity privacy of the communicators.Through analyzing anonymous communication traffic,the association between anonymous users and the websites they visit can be identified.And the limited number and sensibility of Tor hidden service websites make them more vulnerable to website fingerprinting attacks.Therefore,the fingerprinting attack in anonymous communication and the technology to prevent it have become one of the hotspots of research on anonymous communication security.Website fingerprinting attack is a kind of traffic analysis attack against the web side.It is able to identify the target website visited by users using the information of request and response packets.The upgrade of website fingerprinting defense methods has revealed shortcomings,such as high training costs and poor robustness,of existing website fingerprinting attack models.Aiming at the concept drift of website fingerprinting attack and the reduction of recognition rate caused by network traffic filling and confusion,two website fingerprinting attack models based on time series features are proposed.Furthermore,this paper designs a low-cost website fingerprinting defense and implements an adaptive website fingerprinting attack system.(1)In response to the concept drift problem of website fingerprinting attack,a neural network-based website fingerprinting attack model PCTM is designed.This model is conducive to improving the classification accuracy,reaching 98.96% in the closed world.It can still achieve a recognition accuracy of more than 86% when faced with fingerprint data after six weeks of training,effectively alleviating the problem of concept drift.Aiming at the problem that the accuracy of the attack model decreases obviously under the influence of website fingerprint defense,a temporal feature extraction algorithm suitable for website fingerprinting is designed,and a website fingerprinting attack model ETS-WF based on extended time series is proposed.By virtue of using the differential information between the website fingerprint sequences,the comparison data between the obfuscated fingerprint and the original fingerprint replaces the original fingerprint sequence to effectively deal with various website fingerprint defenses.Under the influence of non-delayed defense methods,it has achieved an accuracy rate of more than 90%,which is more than20% higher than the current level and is also improved by more than 10% in the face of delayed defense methods.(2)Using the longest common subsequence algorithm,this paper designs a new website fingerprinting defense method.For each target fingerprint,it uses multiple similar real fingerprints for obfuscation and controls the defense cost.This can effectively fight against multiple website fingerprinting attacks.The results show that the website fingerprinting defense designed in this paper achieves more than most current fingerprinting defenses with less than 20% bandwidth cost and 0% time cost.At the same time,this substantially reduces the implementation cost compared to comparable level of defenses.(3)Using the proposed website fingerprinting attack method,this paper designs and implements a new adaptive website fingerprinting attack system.It integrates data preprocessing,pre-classification,website fingerprinting attack model training and result analysis.With the adaptive attack system,it simplifies the deployment process of website fingerprinting attack.At the same time,it improves the accuracy of website fingerprinting attack in realistic scenarios from less than 50% to more than 80%.