A new steganographic method resisting statistical attack

In this paper we present a new steganographic method for JPEG images that can resist both chi-square attack and Fridrich's "Breaking the F5" statistical attack. A key element in our method is a compensatory strategy that keeps the first-order statistical property of DCT coefficients in the stego-image almost identical with that in the cover-image, thus the statistical attacks are defeated. Unlike the correcting transform adopted in OutGuess that modifies steganographically usable coefficients to correct the statistical deviations caused by the embedding process leading to a relatively small steganographic capacity, the compensatory scheme in our method uses steganographically unusable coefficients (coefficients with absolute value equal to zero) to compensate the distortion of the first-order statistical property of the stego-image induced by the embedding process. Therefore both the steganographic capacity and the quality of the stego-image remain high. Experimental results based on a large set of test images demonstrate the performance of our method in the aspects of both security and capacity issue.