Research On Steganalysis Based On JPEG Images

The illegal use of steganography seriously damaged the network information security. How to effectively prevent the illegal application of steganography has been important component of national security. JPEG format is contained in the spread of the current density on the network main image formats, so the relevant steganalysis algorithm is also very large, but most of which have good results only for a particular type of steganographic algorithms. The research of a variety of effective general detection algorithm for steganography algorithms is the current hot and difficult technology. To counter the steganalysis by adopting various means to disrupt the statistical pattern, steganography algorithms continue to improve and develop. Old detection algorithms are often far from ideal for the improved algorithm in the results. One typical example is that the proposed detection algorithm in the past open literature for the PQ and the F5 algorithm performance are poor. This paper for the PQ and the F5 steganographic algorithm design simple and effective detection algorithms respectively.The main research findings of this paper includes:(1) Discussing the indexes and classification of steganalysis, analysing the general JPEG steganographic algorithms in depth and specificly,comparing various types of universal steganalysis algorithms and special steganalysis algorithms,analysing their advantages and disadvantages, making preparation for the main work of this article.Discussing the PQ steganography method which proposed by J. Fridrich and directing at compressed JPEG images,Current PQ steganalysis result are not always effective. According to impact on the AC coefficients of the JPEG image which caused by PQ steganography,this article introduces the low-frequency AC coefficients difference and AC coefficients shrinkage characteristics as two types of statistics. Extracting the features from the stego JPEG image, and using the Fisher linear classifier as the training and testing tools to determine whether the images had been steganographic,which making detection rate rise up to 85% or more.Proposing a detection algorithm for the F5. This algorithm on the one hand direct at the statistical impact on the image closely, and get the centroid change value of DCT coefficient and energy value of AC coefficients as feature vectors; the other hand, use the strong learning ability of the SVM classifier. Finally, contrasted with the detection algorithm proposed by J. Fridrich for the F5 and comparative experiments show that in the case of a low false alarm rate,the algorithm can maintain a higher detection rate despite of the various embedded rate into stego image, And the algorithm runs faster, and highly practical.