Research On Video Steganalysis Algorithm For H.264/AVC

Steganalysis is a kind of opposition or attack technique of steganography, and its main function is to detect and analyze whether the carrier contains the secret information. Steganography is the study of information hiding which is to embed secret information in the mass media and other carriers to achieve covert communication and it's not easily found in third-party. The existing steganalysis algorithms are mostly designed for specific known steganography algorithms. There are at least two problems. On the one hand, effective steganography algorithms usually have invisibility and security. Majority of steganography algorithms are not published, such as patents, etc.; On the other hand, some new steganography algorithms are constantly being proposed. So it is not effective to find most of the stego carrier(such as pictures, videos, etc.) owing to the corresponding specific steganalysis algorithms for the few open steganography algorithms. Therefore, under the condition of unknown steganography algorithms, steganalysis become an important challenge by analyzing the common differences of before and after the video, and it is called blind steganalysis.A new video steganalysis algorithm based on weighted undirected graph was proposed. First, the co-occurrence matrix reflected the correlation between pixels, and the correlation between the frames was reflected by the weighted undirected graph, which can better reveal the macroscopic differences between frame-to-frame and pixel-to-pixel. Experimental results showed that the algorithm using weighted undirected graph can effectively discriminated the stego video and the original video, and it has a high accuracy rate.A new steganalysis algorithm was proposed based on the Markov features of utilizing the temporal correlation among video frames, which was modeled as a special stochastic process. First, divide all frames of a video into a new group called the P-I-P frame structure. Then, calculate the co-occurrence matrix of pixel-to-pixel to get the most similar macroblock among adjacent frames and build the Markov model between the most similar macroblocks. Finally, we tried to conduct the blind steganalysis to further determine whether any secret information was embedded in the video according to the changes of correlation by P-I and I-P frame structures before and after the embedded information videos. The experimental results showed that the new steganalysis algorithm based on the Markov features can achieve the desired effect, and it is conducive to further analysis and research.