Identification of Encrypted Video Streaming Based on Differential Fingerprints
Hua Wu, Zhenhua Yu, Guang Cheng, Shuyi Guo
Abstract
Nowadays, because of the convenience and broad reach of social media platforms, such as YouTube, Facebook, and Twitter, terrorists have been taking advantage of these platforms to spread illegal videos. Existing researches extract video fingerprints based on statistical data and inevitably overlook some features, which makes it impossible to accurately represent the encrypted video streaming. Besides, almost no research can identify the target video from a large video dataset on the premise of high accuracy and low false-positive rate. Therefore, in this paper, we extract differential fingerprints to represent the encrypted video streaming more accurately and identify the encrypted video streaming using our local sequence alignment algorithm. We conduct our method on a large dataset containing 200,000 videos. The results show that by continuously sniffing the encrypted video streaming for 12 seconds, the precision of our method can exceed 98%, and the false-positive rate is nearly 0.