Research On Video Steganalysis

The security of information hiding used as covert communication technique is an urgent problem in the filed of information security. The goal of steganalysis is to detect the presence of cover data, to estimate the embedded message length, and finally to extract the hidden message. Steganalysis can effectively detect and monitor the illegal information in order to ensure the security of network. Therefore, it becomes the hot spot in the field of information security. In this dissertation, research efforts are concentrated on digital video research. The embedding characteristics and essence of the video information hiding algorithms are analyzed and revealed. Further, the video steganalysis is studied in-depth. The main work is as follows:(1) A novel video steganalysis is proposed against the spatial spread spectrum steganography. Under an independence assumption, the aliasing effect is revealed, which is caused by embedding. At the same time, the wavelet filter and decorrelation techniques are used to highlight the aliasing effect. The proposed algorithm can effectively detect the secret message with different embedding strength in video with different compression bitrates by measuring the aliasing degree.(2) A video steganalysis called EOB-MC is proposed against differential energy watermarking algorithm, which embeds bits in compressed videos by selectively discarding high frequency DCT coefficients in certain image regions. The end of block tag (EOB) can be moved forward by the embedding algorithm. Based on this characteristic, EOB-MC steganalysis algorithm exploits temporal statistics of video sequence to estimate a statistical model and utilizes the K-S hypothesis test to detect the presence of the hidden message.(3) A video steganalysis technique is proposed against the motion vector video steganography methods. Through analyzing the influence introduced by the embedding process, the statistical properties sensitive to embedding are extracted as a feature vector. The support vector machine is then used as a discriminator. Experimental results show that the proposed method can effectively detect the embedded data introduced by three different motion vector steganography algorithms.In addition, the dissertation also researches other aspects of video steganalysis. An energy distribution model is set up to detect the presence of the secret message which can be embedded in the specific region of DCT coefficients matrix. Besides, the MSU Stego Video steganography software is analyzed and two steganalysis methods are designed.All in all, the dissertation provides an in-depth investigation into video steganalysis including spatial steganalysis, DCT coefficients steganalysis and motion vector steganalysis. The experiments results demonstrate that the proposed steganalysis algorithms are effective.