Research On Key Issues Of Quantitative And Locating Image Steganalysis

Quantitative and locating steganalysis are the hot and difficult issues in the field ofsteganalysis. Currently, there have been numerous excellent results on quantitative and locatingsteganalysis, but which still can not satisfy the requirements of steganalyzers. This thesis mainlydiscusses the researches on quantitative and locating image steganalysis, includes seven chapterswhich can be summarized into the following three parts:The first part: background knowledge and state of the art. As viewed from the applicationand technical background, the practical and theoretical values of steganalysis are described. Theresearch review of digital steganography and steganalysis are introduced briefly. The state of theart in quantitative and locating image steganalysis are surveyed in detail. And some key issues onquantitative and locating image steganalysis are pointed out.The second part: the quantitative image steganalysis. In regard to the problem of estimatingthe low embedding ratio of MLSB (Multiple Least Significant Bits) replacement steganography,two quantitative steganalysis methods are respectively proposed for TMLSB replacement(Typical MLSB replacement) steganography and IMLSB replacement (Independent MLSBreplacement) steganography based on the transition relationship among pixel groups; in regard tothe problem of estimating the high embedding ratio of MLSB replacement steganography, twoquantitative steganalysis methods are respectively proposed for IMLSB replacement and ID-MLSB replacement (Independent MLSB replacement with Different ratios) steganography basedon the weighted stego image; in response to the problem of measuring the difference betweenimage features, based on relative entropy, a method is proposed to compute the differencebetween histogram-like features, then based on which, a quantitative steganalysis method isproposed for JPEG image steganography.1. Quantitative steganalysis of MLSB replacement steganography based on pixel grouptrace. First, the adjacent pixel groups are used to described an image and the transitionrelationship among pixel groups are given based on the pixel group trace set and trace subset ofMLSB replacement; second, based on the change of the pixel group’s noise level when applyingtwo categories of symmetrical masks, some statistical characteristics of the cover and stegoimages are pointed out; third, based on above transition relationship and statistical characteristics,two quantitative steganalysis methods are proposed for TMLSB replacement and IMLSBreplacement steganography respectively. Experimental results show that the proposedquantitative steganalysis methods significantly outperform other methods for low embeddingratio.2. Quantitative steganalysis of MLSB replacement steganography based on weighted stegoimage. First, based on the weighted stego image of IMLSB replacement, a quantitativesteganalysis method is proposed for IMLSB replacement, and the relationship between theproposed method and the weighted stego image steganalysis method for TMLSB replacement ispointed out; then, based on the weighted stego image of ID-MLSB replacement, a quantitativesteganalysis method is proposed for ID-MLSB replacement steganography. Experimental results show that when the embedding ratio is high, especially when the embedding ratio is close to1,the proposed quantitative steganalysis methods own significantly smaller error than others.3. Quantitative steganalysis of JPEG image steganography via measuring the differencebetween histogram-like features based on relative entropy. First, based on the optimum twohypotheses test---likelihood ratio test, a relative-entropy-based method is proposed to computethe difference between histogram-like features; then, based on the proposed differencecomputing method and the idea of machine learning, a quantitative steganalysis method isproposed for JPEG image steganography via measuring the difference between histogram-likefeatures based on relative entropy. Experimental results show that for JSteg, F5and somevariants of F5, such as-F5and nsF5, compared with the typical quantitative JPEG steganalysismethod which directly computing the diffence between two histogram-like features, the proposedquantitative JPEG image steganalysis can estimate the modification ratio with smaller error, andwould be affected by the actual modification ratio less.The third part: the locating image steganalysis. For the case of owning multiple stegoimages which are embedded message into along the same embedding path, a general locatingsteganalysis method is proposed based on sample block selection and quantitative steganalysis;for the case of owning single stego image of TMLSB replacement, a property of TMLSBreplacement is analyzed based on the weighted stego pixel, and a locating steganalysis methodfor sequential TMLSB replacement steganography is proposed based on minimum sumsubsequence; for the problem of stego key recovery, from the idea of collision attack, a reversingmethod is proposed based on locating steganalysis.1. Locating steganalysis based on sample block selection and quantitative steganalysis. Forthe case of owning multiple stego image with the same embedding path, first, the problem ofdetermine the stego positions is converted into estimating the embedding ratios in the positionsfor determining; second, based on sample block selection, a method is proposed to estimate theembedding ratio in each position; third, based on the estimated embedding ratio, a generallocating steganalysis is proposed; finally, based on the general method, some locatingsteganalysis algorithms are proposed for LSB replacement steganography and TMLSBreplacement steganography. Theoretic analysis and experimental results show that the proposedalgorithm not only contains Ker’s locating steganalysis algorithm for LSB replacement, but alsocan be to locate the stego pixels of TMLSB replacement; and with the increase of the number ofstego images with the same embedding path, the exactness of the proposed locating steganalysisalgorithms will be improved significantly.2. Locating steganalysis of TMLSB replacement based on minimum sum subsequence. Forthe case of owning single stego image of TMLSB replacement, first, the problems necessary tobe settled when applying the idea of weighted stego image to the locating steganalysis ofTMLSB replacement are pointed out; second, a property of TMLSB replacement is analyzedbased on weighted stego pixel; third, based on the property, the process of estimating the startand end positions of sequential TMLSB replacement is converted into the problem of findingminimum sum subsequence, and the locating steganalysis algorithms for sequential TMLSB replacement are propsoed. Experimental results show that the proposed locating steganalysismethod for sequential TMLSB replacement can estimate the start and end positions with highaccuracy. Additinally, the proposed method contains the Ker and B hme’s algorithms forsequential LSB replacement, owns wider applicability.3. Stego key recovery based on locating steganalysis. First, from the idea of collision attack,a stego key recovery method is proposed based on locating steganalysis; second, based on thenormal approximation of binomial distribution, the effects of false alarm rate, miss alarm rate,embedding ratio and accuracy of locating steganalysis on the efficiency of stego key reversing isanalyzed; third, a stego key recovery algorithm for LSB replacement is given based on theimproved weighted stego image method, and a steg key recovery algorithm for TMLSBreplacement is given based on the weighted stego image method. Experimental results show thatthe proposed method can recover the correct stego key with higher ratio and improve the speedof recovery. Additionally, the proposed stego key recovery method can utilize the results oflocating steganalysis to recover the stego key for more steganography.Finally, a conclusion with a discussion of the direction for the future research is given.