Blind detection for JPEG steganography

In this paper, a blind steganalytic scheme is proposed to effectively detect JPEG steganography. Nine statistical models are constructed from the DCT and decompressed spatial domain for a JPEG image. By calculating the histogram characteristic function (HCF) and the center of mass (COM), we measure the energy distribution of each model as one part of our feature set. Moreover, for each model constructed from the DCT domain, we select a suitable set of individual elements which may hold main information of the model as the other part of our feature set. Support vector machines are utilized to construct classifiers. We report results for some currently popular JPEG steganography (Jphide [8], Outguess [9], Steghide [10], MB1 [11], and MB2 [12]). Experimental results demonstrate that the proposed scheme provides the best detection accuracy comparing with the similar steganalytic schemes reported recently.