Single Band Statistics and Steganalysis Performance

Most universal steganalysis techniques use an image model to reconstruct an estimate of the original, unmanipulated cover from the input. Differences between reconstructed and input images are an indication of a steganographic manipulation. In this paper, we analyze the relation between the modeling error of the image model and detection performance in the wavelet domain. Based on the modeling error we define a measure of separability which is highly correlated with detection performance. We find that in uncompressed images only fine scales play a role in steganalysis, whereas separability is spread out over all scales in JPEG images.