The Steganalysis Of Perturbed Quantization Like Steganography

Digital image steganography and steganalysis are important branches of the information hiding, after almost twenty years’development, the conflict between them is more and more serious, which has been a main issue in the information security field. In the past few years, some steganography methods with higher security are poposed, such as perturbed quantization like steganography, which minimize embedding distortion. So they can not be detected effectively by the typical steganalysis methods, it has brought great difficulties and challenges to steganalysis. This thesis mainly discusses the researches on the steganalysis of perturbed quantization like steganography, and the contents are as follows:1、The principle of perturbed quantization like steganography is analyzed. The analysis of steganography principle is crucial for sensitive feature extaction, feature comparison and reliable detection algorithm design. Based on the changeable coefficient selection rule, the relationship of contributing position and the second quantization step is given; then there is a discussion about the value of changeable coefficients in cover and stego images, on this basis the statistical laws of local histograms are analyzed with a conclusion that the coefficient histograms of contributing positions will be changed by embedding.2、For the problem that the research on steganalysis of perturbed quantization like steganography is lack of a universal detection framework, a framework with strong currency is proposed, which contains three parts:the image database construction, the sensitive feature extraction, the classifier design and train, the second part is the most important part and consist of two schemes:sensitive feature extraction based on the analysis of existing features and sensitive feature extraction based on the analysis of steganography principle.3、For the problem that the existing steganalysis of perturbed quantization like steganography are lack of theory analysis of feature extraction and comparison, a detection algorithm based on improved co-occurrence matrix is presented. Drawing on the experience of existing co-occurrence matrix feature, the improved feature is obtained by eliminating the invalid charater statistics, and then the conclusion that the improved feature is more sensitive for perturbed quantization like steganography is found with the theory analysis, on this basis the detection algorithm based on improved co-occurrence matrix is presented. The conclusion about feature comparison is verified by the experiments.4、For the problem that the existing steganalysis do not fully consider the particular modification made by perturbed quantization like steganography, which lead to the feature is non-sensitive, a detection algorithm based on histogram difference in contributing position pairs is presented. On the basis of analysis of steganography principle, the histogram difference in contributing position pairs is analyzed with the conclusion that it has some potential on capturing the changes by embedding, so it can be extracted as feature for the steganalysis, then the detection algorithm based on histogram difference in contributing position pairs is presented. Experimental results show that the presented algorithm has higher detection accuracy comparing to the current typical steganalysis methods.Finally, a conclusion of this thesis is given and some other research contents to be studied are pointed out.