Blind image steganalysis based on reciprocal singular value curve

In this paper, a new SVD-based feature set is introduced for steganalysis both in spatial and JPEG domains. Previously, reciprocal singular value curve has been used for no-reference image quality assessment. In fact, embedding secret messages in steganographic approaches is similar to adding some weak noise to the original media. Hence, the disturbance of natural image statistics is explored to extract the feature vector for steganalysis. In the proposed scheme, the alternation of singular value curve is utilized for constructing the steganalysis feature vector. The experimental results illustrate an acceptable performance of the proposed feature in universal steganalysis.