Research On JPEG Image Steganalysis

With the rapid development of information hiding technology, on the one hand, it is easy to use some steganography tools to convey secret message, on the other hand, the abuse of these tools is a serious threat to the security of network information, so it is urgent to study steganalysis technology against these threats.This paper first introduces the basic theory and the development status of steganalysis. Then, two steganalysis schemes are included, and the research results are as follows:(1) A blind steganalysis based on regional correlation and second-order Markov transition probability matrix is proposed for JPEG images. By analyzing the region correlation of JPEG image, the Markov transition probability matrix is used to capture the correlation of intra-block and inter-block DCT coefficients. In addition, the calibrated features are extracted from the calibrated images. The difference features between the original and calibrated images are used for training and classifying. Three different image libraries are used to detect the five kinds of typical JPEG steganography schemes. Experimental results show that, in comparison to the several effective steganalysis, the proposed scheme improves the detection accuracy on some JPEG-based steganography schemes, including outguess, Steghide and MB1.(2) Aiming at high detection accuracy for JPEG images in a low embedding ratio,a blind steganalysis is proposed. Analyzing the correlations of intra-block and inter-block of JPEG image, we can obtain the modified Markov features and joint probability density features from DCT domain and DWT domain, and extract the second order Markov transition probability density from spatial domain. The original features and differences between the original and the calibrated are employed to train and classify. Three image libraries and five kinds of typical JPEG steganographic algorithms are conducted in our experiment. Experimental results have shown that,compared with several existing steganalysis algorithms, our scheme has higher detection accuracy for JPEG images in a low embedding ratio.