Font Size: a A A

Image Steganalytic Feature Analysis And Steganographic Algorithm Recognition

Posted on:2015-12-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:J C LvFull Text:PDF
GTID:1108330482479237Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
The researches on image steganalytic feature analysis and steganographic algorithm recognition are of great theoretical value and practical significance for improving the steganalytic accuracy and generalization, extracting confidential messages and so on. This dissertation mainly focuses on some key issues of image steganalytic feature comparison and selection, feature dimensional reduction and steganographic algorithm recognition. The dissertation includes eight chapters and could be divided into three parts as following:The first part: background and state of the art of corresponding researches. With respect to the practical application and technical research, the practical significance and theoretical value of steganalytic feature analysis and steganographic algorithm recognition are discussed. The corresponding concepts and researches of steganography and steganalysis are introduced. The research progresses of image steganalytic feature analysis and steganographic algorithms recognition are stated in detail.The second part: The researches on image steganalytic feature analysis. A method for steganalytic feature comparison and selection is proposed based on the feature changing rate, and also a method for feature dimensional reduction is proposed based on Fisher criterion. On the basis of those, two types of typical JPEG steganalytic features co-occurrence matrix and Markov transition probability matrix are compared for selection, and the reduction of typical spatial domain steganalytic feature SPAM(Subtractive Pixel Adjacency Matrix) and typical JPEG steganalytic feature CCPEV(Cartesian Calibration based feature proposed by PEVny) are analysis. The details are as follows:1. The feature analysis method. For the performance comparison and selection of steganalytic features, the changing of the feature before and after steganography are analyzed and the feature changing rate is taken as the sensitivity of feature to steganography, then, a method for feature comparison and selection is proposed based on the feature changing rate. For the component selection and dimensional reduction of high-dimensional steganalytic feature, the separability of single feature component and multiple feature components are evaluated based on Fisher criterion according to the principle of “the within-class distribution is close and the between-class distribution is scattered”, then, a method for feature dimensional reduction is proposed based on Fisher criterion.2. Comparison of typical steganalytic features. For two types of typical steganalytic features co-occurrence matrix and Markov transition probability matrix, when they are used to detect JPEG steganography that make the histogram shrinking(for example: F5) and preserve the histogram(for example: Outguess and MB1), the sensitivities of these two features two steganography are analyzed and compared based on the proposed feature comparison method. Then, the comparison results for performance are obtained, and new features are obtained by fusing excellent feature components. Experimental results show that, the detection performance of the newly fused feature are superior to existing typical features, and indicate that the proposed analysis method is reasonable and effective.3. Dimensional reduction for typical steganalytic features. Using the proposed Fisher criterion based feature dimensional reduction method, and according to the principle of “within-class distribution is close and between-class distribution is scatter”, the spatial domain typical steganalytic feature SPAM and JPEG domain typical steganalytic feature CCPEV are analyzed and reduced for blind steganalysis, respectively. At the same time, when the feature CCPEV is used to classify JPEG multi-class stego images, corresponding dimensional reduction algorithm is presented. The experimental results show that, under the condition of preserving the steganalytic accuracy, the proposed feature dimensional reduction method can effectively reduce the feature dimension, and can also improve the steganalytic efficiency greatly.The third part: Research on the recognition of image steganographic algorithm. A method for steganographic algorithm recognition is proposed based on the identifiable statistical feature(IDSF) for the specific steganography. On the basis of that, the IDSFs of multiple types of typical steganographic algorithms in spatial domain and JPEG domain are proposed, respectively, and the corresponding recognition algorithms are also presented. The details are as follows:1. A general method for steganographic algorithm recognition. According to the special modifications of the specific algorithm to image data during embedding, extract sensitive features that can capture the special modifications by constructing proper feature extraction sources, and take them as IDSF of the specific steganography distinguished from others. On the basis of that, a steganographic algorithm recognition method based on IDSFs is proposed in combination with classifier training.2. Recognition of typical JPEG steganographic algorithms: JSteg, PQ(Perturbed Quantization) and F5-like steganography. According to the phenomenon that the symmetry of coefficients histogram is destroyed by JSteg embedding, the IDSF of JSteg is extracted based on symmetry difference of coefficients histogram. According to the characteristic that PQ will not embed messages in some fixed parts of the DCT coefficients, the IDSF of PQ is extracted based on the statistical dependency difference of neighboring DCT coefficients. According to the characteristic that F5-like steganography subtract 1 from the absolute coefficients, the IDSF of F5-like steganography is extracted based on the influences of neighboring coefficients with different positive and negative cases to statistical properties of both spatial and frequency domains. The experimental results show that, based on the extracted IDSFs and the proposed steganographic algorithm recognition method, the steganographic algorithms of JSteg, PQ and F5-like steganography can be reliably recognized from multi-class stego images. Although some types of stego images are unknown, the proposed algorithms can still achieve well classification accuracy.3. Recognition of typical spatial domain steganographic algorithms: 3LSB(Least Significant Bit), 2LSB and LSB substitution algorithms. According to the characteristic that substitution steganography will change the statistical properties of multiple-set of pixel pair, the IDSFs of 3LSB, 2LSB and LSB substitution steganography are extracted based on pixel deviation or shifting pixel deviation. The order to recognize them is analyzed according to the steganographic characteristic, and the corresponding recognition algorithms are presented. Experimental results show that, based on the extracted IDSFs and recognition method, the steganographic algorithms of 3LSB, 2LSB and LSB substitution can be reliably recognized from multi-class stego images. Although some types of stego images are unknown, the proposed algorithms are still effective.Finally, a conclusion is given and the future researches are also discussed in the thesis.
Keywords/Search Tags:Steganalysis, Feature comparison, Feature selection, Feature dimensional reduction, Steganographic algorithm recognition, Identifiable statistical feature(IDSF)
PDF Full Text Request
Related items