TLD-WF: Surfing Freely Under Website Fingerprinting Attacks

Abstract

Anonymous networks hide communication content and communication relationships by encrypting the content and transmission path of traffic. But the website fingerprinting attack can determine the websites a user visits by analyzing the user’s traffic patterns and website fingerprints without decrypting the traffic. To defend against website fingerprinting attacks, most of the existing methods reduce the accuracy of analysis by increasing the access delay and background traffic. Unfortunately, such methods not only degrade the user experience, but also affect the performance of anonymity networks. By adopting the Tor pluggable transmission technology, this paper proposes a new targeted, lightweight defense to against website fingerprinting attack (TLD-WF). TLD-WF uses the TLD-flow module to obtain the analyzed flow information with an accuracy rate of over 90%, and conducts targeted dynamic scrambling. TLD-WF not only effectively guarantees the anonymity and security of Tor users, but also has the advantages of zero latency overhead and low bandwidth overhead. The experimental results show that TLD-flow can reduce the analysis accuracy of the CNN classifier from 93.2% to around 8.3% under normal conditions. With guaranteed zero latency and 45% bandwidth overhead, the analytical accuracy of the CNN classifier will be reduced by about 50%.