Blind image steganalysis based on wavelet coefficient correlation

Abstract

To detect the presence of information in a stego image more reliably, a blind JPEG steganalysis method based on inter- and intra-wavelet subband correlations in the wavelet domain is proposed. First, after two-level wavelet decomposition, the joint probability density of each subband’s difference from neighboring coefficients in the horizontal, vertical, and diagonal directions is calculated, and the entropy and energy are extracted from the joint probability density matrix as features. Then the image is decomposed into three subbands, and the PDF (probability density function) is extracted from each subband’s wavelet coefficient. Finally, the three kinds of features described above are combined to detect the image. In experiments, the proposed method is compared with various other blind steganalysis methods, and the impacts of different feature combinations on detection accuracy are discussed. Experimental results from typical JPEG image stego algorithms such as F5, Jsteg, Outguess, and Jphide show that the proposed method significantly outperforms typical blind steganalysis methods. The proposed method also has some detection capabilities for double-compressed images.