Research On Encrypted Video Channel Identification Based On Traffic Fingerprin

The time-sync comments have been prevalent in modern live streaming services to provide a real-time interaction experience for viewers.However,similar to the video traffic,the statistical features carried by time-sync comments can also leak the key information of the video channel and lead to potential risks of privacy leakage after analysis.Most video platforms use transport layer security protocol(TLS)to encrypt the video traffic,which will make it difficult to obtain the unencrypted video content through traffic monitoring,so the previous video channel identification strategies usually make fingerprints based on video bitrate and the statistical features of video traffic.But most of them presume strict requirements on the implementation environments,which often assume that there is no interference from irrelevant traffic flows or bandwidth fluctuation,and the identification accuracy will decrease with the destruction of the fingerprint when the network condition become worse.However,the time-sync comments are distinct from video traffic flows and has more obvious features,which has higher adaptability to the complex network condition and easier to leak the privacy of the viewers even if there are noise interferences in the target flow.Therefore,the main contributions of this paper aiming at the video monitoring in complex network conditions are as follows:(1)This paper analyzes the noise interference in the real network conditions from the viewers,and a live channel identification system based on encrypted time-sync comments is proposed.The system can effectively identify video channels even if the viewers in a complex network condition with noise interference such as bandwidth fluctuation.Firstly,a filter module based on TLS label and convolutional neural network(CNN)is proposed to filter the irrelevant traffic.Further,a comment rate estimation method is developed through investigating the relationship between comment number and packet length.Finally,the dynamic time warping(DTW)algorithm is improved for similarity matching in delay tolerant environment,and the features generated by the above steps is identified by support vector machine(SVM).(2)The above scheme needs to monitor the victim for a long time to achieve higher accuracy,and SVM cannot be embedded into the traffic filter model for end-to-end training.To solve these problems,scheme(2)analyzes the noise interference on bitrate-based video fingerprints,and a comprehensive identification scheme is proposed based on time-sync comments and video traffic.Firstly,a traffic feature extraction algorithm based on longest common subsequence(LCS)is proposed to extract features from mutilated fingerprints.Then,a neural network model developed with attention is used to identify the live channel with video features and time-sync comment features.The experimental results show that the identification accuracy of scheme(1)can reach 92.1%by monitoring the time-sync comments from Youtube live channels for 200 seconds,and the same accuracy can be achieved by monitoring the traffic for only 70 to 80 seconds after introducing the video bitrate features mentioned in scheme(2),which is far less than the time required by other identification methods in the latest research.