FD-WF: A Multi-tab Website Fingerprinting Attack Based on Fixed Dimensions for Tor Network
Ruizhe Zhang, Shang-Nan Yin, Jinfu Chen · 2024
Website Fingerprinting Attack (WFA) is an effective method of network monitoring, which analyzes network traffic to identify the specific website or web page that a user is browsing. The performance of previous WFA, which assume singletab scenarios, deteriorates significantly in real-world multi-tab environments. While this issue has been acknowledged and some research has been conducted on multi-tab WFA, limitations in the accuracy and efficiency of multi-tab classification still persist. Specifically, extended research on mixed-tab scenarios, where the number of tabs is unknown, still lacks sufficient attention. In this paper, we propose FD-WF, a multi-tab WFA model based on fixed dimensions on the Tor network. This model mitigates the issue of blurred features in single-tab images caused by the expansion of multi-tab mixed traffic sequences. It enhances the ability to accurately identify and classify multiple web pages users are browsing. We propose a minimum padding optimization function to improve the performance of fixed-dimension image generation and introduce an enhanced ResNet-18 model to better address the fingerprint classification challenge for multi-tab scenarios. The experimental results verified the feasibility and effectiveness of the proposed method. While ensuring the performance of singletab WFA, our model achieved a F1 Score of 96% for multi-tab web pages. Even in the more complex mixed-tab scenario, we still achieved an F1 Score of 78%.