Accurate Identification of Encrypted Videos in Asymmetric Routing Scenarios

Encrypted video transmission protects users’ privacy, but this also makes it more challenging to identify harmful videos on the Internet. Existing methods are based on bidirectional traffic and can not be applied in wide areas where asymmetric routing is widely available. Moreover, few can identify encrypted videos accurately when applied to large video databases. In this paper, we designed a method to accurately extract video stream features from the unidirectional flow and identify the encrypted videos by constructing Markov models. We implemented our method on the extensive dataset containing 470,398 video fingerprints. The experimental results show that the accuracy can reach 98.21%.