Real-time steganalysis for streaming media based on multi-channel convolutional sliding windows

In recent years, covert communication technologies based on Voice over Internet Protocol (VoIP) have received more and more attention, which meanwhile poses a significant threat to the security of cyberspace. In this paper, we are chiefly concerned with improving the accuracy and efficiency of detection of covert communications, and we propose a real-time VoIP steganalysis model to tackle these issues. Multi-channel convolutional sliding windows (CSW) are developed to analyze the correlations between a given frame and its neighboring frames in a VoIP signal. Within each sliding window, we employ two feature extraction channels to extract correlation features from the input signal. Each channel is constructed of multiple convolutional layers having a large number of convolution kernels. The extracted features are then fed to a forward fully connected network for feature fusion. By analyzing the statistical distribution of these features, the discriminator will determine whether the input speech signal contains covert information or not. We designed several experiments to test the proposed model's detection performance under various conditions, including different embedding rates, different speech lengths, etc. Experimental results show that the proposed model can efficiently and accurately detect steganographic voice streams, especially in the case of low embedding rates. In addition, further experiments demonstrate that the proposed model can attain nearly real-time detection of VoIP speech signals and achieve state-of-the-art performance. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

This paper proposes a real-time VoIP steganalysis model to improve the accuracy and efficiency of detecting covert communications. The model utilizes multi-channel convolutional sliding windows and statistical feature analysis to achieve high detection performance, especially for low embedding rates.