Steganalysis For JPEG Image Based On SVM

Digital image steganography is one kind of covert communication technology,which embed secret information into a public digital image to hide and transferinformation. Being used in military, commercial and other fields, it brings a newcovert communication method for the community and a new threat at the same time.Digital image steganalysis is a warfare technology of digital image steganography thatbased on the method of statistical analysis to detecet image steganography andconfirm the existence of the secret information in the digital image so as to achievethe purpose of monitoring and destroying covert communication.While steganography and steganalysis oppose each other, they also promotemutually, receiving wide attention in the information security field of digitalmultimedia. Because of the widespread of JPEG images on the network, the JPEGimage steganalysis has a good application prospect. The F5algorithm is widelyrecognized as a higher security JPEG image steganography algorithm, so a targetedsteganalysis have important practical significance. Focusing on the JPEG imagesteganalysis, the main work of this paper is summarized as follows:A universal JPEG images steganalysis method based on second-order statisticalcharacteristics of the DCT coefficients is reported. Firstly, this method extract50-dimensional feature vectors which be made up by co-occurrence matrix featuresand differential characteristics from three JPEG image steganographies (Jsteg,OutGuess and F5), and then use the advantages of support vector machine (SVM) inthe nonlinear and high dimensional pattern recognition to conduct steganalysisexperiments. The experimental results show that the detection effect of JPEG imagesteganography (including the correct detection rate, false detection rate and the virtualseizure rate) is better than the existing methods.A steganalysis algorithm against F5is reported. The steganalysis algorithmbased on the contractionary effect result in DCT coefficient histogram of imageembeded information by F5extracts11dimensional feature vectors and use supportvector machine to conduct related experiments, showing the proposed algorithm caneffectively detect F5steganography algorithm.The two steganalysis methods reported in this paper have lower featuredimension and computational complexity and the correct detection rate is relativelyhigh.