This repository contains the code for the paper "Unveiling the Unseen: Video Recognition Attacks in Complex Network Environments and Playback Modes". As the content of this paper involves sniffing attacks, to prevent misuse, we have limited the functionality of the open-source program.
We only provide executable programs and restrict the dataset to data prior to March 2024, providing only five traffic cases under four conditions. The program is divided into Windows and Linux versions. data.cfg contains the configuration related to file paths; twitter is the program for restoring video segment lengths, make_db is the program for building the database, and match_mul_db is the recognition program.
The dataset is too large, so it is placed in the release version. The twitter_fingerprint_current.csv in the data folder is the fingerprint of the video corresponding to 20 video streams, and twitter_fingerprint_large-sal.csv is the fingerprint of 302,358 videos. The folders are divided into under a fluctuating network environment with seeking playback, under a fluctuating network environment with sequential playback, under a stable network environment with seeking playback, under a stable network environment with sequential playback, and all folder containing the above four cases.
The program execution flow is to first execute twitter to generate the length of the restored video segment, then use make_db to generate the fingerprint library, and finally use match_mul_db for fingerprint recognition. The recognition result is displayed in 0_mul_DB_match.
The experimental environment of this paper's program is win11 (16GB memory) and Ubuntu22.04 (16GB memory).