Research On Steganalysis Based On Image Content

As a quite important technical tool for image information security, image steganalysis has become an attractive hotspot of the multimedia information security to researchers all over the world. Image steganalysis is the opposite technology against steganography, which aims at detecting, extracting, restoring and destroying the secret messages embedded into the cover images. The basic concept of the current image steganalysis is analyzing the embedding mechanism and the statistical changes of the image data caused by secret message embedding. However, most existed steganalysis methods ignore the inherent characteristics of images, and the embedding changes are not correlated with the steganography methods, but also with the local statistical characteristics. As a result, the steganalysis performances are highly depending on the image content, and it is significant to study the steganalysis combining with image processing technics and make a research on steganalysis based on image content. This dissertation focuses on steganalysis based on image content. In this thesis, an image is assumed to be a local stationary Markov source, and on the basis of analyzing the correlation between steganalysis features and image content complexity, several content based steganalysis methods are proposed. The contributions obtained in this thesis can be summarized in the following aspects:1. The theoretical and technical methods of image engineering are studied considering image steganalysis. The correlation between the complexity of the content and the statistical features is analyzed based on statistical methods. The analytical results indicate that the statistical features of images vary with the content, and the complexity of the content can be measured by the coefficients of discrete cosine transform(DCT) and texture features. The images can be segmented into regions with coherent statistic features according to the content complexity, which lays a foundation to the research of content-based image Steganalysis.2. A high resolution real image database is set up, and the images in the database are classified according to the complexity of the content. Several specific and universal steganalysis features, including the smoothness of the histogram, subtractive pixel adjacent matrix, merged DCT and Markov features, are extracted, and the correlation between the steganalysis features and the content complexity is analyzed. The existed content based steganalysis models are introduced and the merits and demerits are summed up by analyzing the experimental results. The conclusion can be drawn that such models can effectively decrease the impact of image content and improve the steganalysis performances. Consequently, the segmenting-based steganalysis model is proposed, in which the given image is segmented into several regions according to the content complexity so that the regions have coherent statistical characteristics. The features are extracted from each sub-image for steganalyzing.3. A JPEG steganalysis algorithm based on image segmentation is proposed. The given images are segmented into several sub-images according to the texture complexity measured by DCT features. The steganalysis features based on Merged DCT feature set are extracted from each sort of sub-images with the same or close texture complexity separately to build a classifier. The steganalysis results of the whole image can be figured out through a weighted fusing process of all categories of the sub-images. Experimental results demonstrate that the proposed method exhibits excellent performance and significantly improves the detection accuracy especially when there exist considerable diversities in image sources and contents.4. Considering the unsatisfactory performance on detecting adaptive embedding methods of low dimensional steganalysis feature set, the segmenting-based model is combined with high dimensional features, and a steganalysis method based on JPEG rich models is proposed. The features of the segmented images are extracted, trained and tested separately, and a weighted fusing process is also introduced to get the detective result. The detection performance is effectively enhanced, and the experimental results also demonstrate that combining the segmented image features and the integrated image features can improve the performance more.5. A steganalysis algorithm aimed at spatial steganographic methods is proposed. The given images are segmented into sub-images according to the texture complexity using quad-tree method. High-order differences joint distribution features are separately extracted from each sort of sub-images with the same or close texture complexity to build a classifier. The steganalysis results of the whole image are figured out through a weighted fusing process. Experimental results performed on several diverse image databases and circumstances demonstrate that the proposed method exhibits excellent performances especially for adaptive embedding methods.6. The spatial low dimensional feature set performs poorly in detecting HUGO method. To solve this problem, a local linear transform(LLT) based steganalysis method is proposed. 20 different LLT masks are utilized to catch the subtle embedding changes. The high-order LLT residuals are defined and three order adjacent distribution Markov features are extracted from each segmented sub-image. Experimental results show that the proposed method can obtain almost the same performances as SRM method with only about half of its feature dimensionality and greatly improved computation efficiency.Finally, the whole dissertation is concluded and the future direction of research is given.