Robustness Study of Website Fingerprinting Based on Feature Fusion

With the ongoing advancement of website traffic analysis technologies, techniques such as website fingerprinting and traffic pattern analysis have become increasingly sophisticated. However, the widespread adoption of encryption and anonymity technologies like Tor has also presented new challenges for network regulation. Current defense mechanisms such as WTF-PAD and Walkie-Talkie have relied on adaptive traffic padding by inserting fake packets intelligently and controlling the timing of bidirectional traffic transmission and reception, effectively reducing the accuracy of website fingerprinting. To address the limitations of these defense mechanisms, a robustness-enhanced website fingerprinting method based on time feature enhancement was proposed. This method incorporated the time dimension into feature learning and combined traffic direction and size features, significantly improving recognition robustness. Experimental results showed that under the WTF-PAD defense, the proposed method achieved a recognition accuracy of 94.34%.