Content-Adaptive Steganalysis via Augmented Utilization of Selection-Channel Information

Abstract

Modern adaptive image steganographic schemes embed secret message into textural regions to make it difficult for steganalytic detection. To overcome the presented challenges, existing steganalytic methods incorporate selection-channel information into steganalytic features so as to improve detection capability. In this paper, we extended the maxSRM steganalytic scheme by better exploiting the selection-channel information in two aspects. On one hand, we processed the embedding change probabilities by highlighting the large probabilities to obtain the so called augmented coefficients. On the other hand, we used the augmented coefficients weighted by the approximated probabilities of occurrence of image residuals for computing co-occurrence matrix in steganalytic features. In this way, we further utilized the selection-channel information and make pixels with high embedding change probability contribute more to final steganalysis features. Experiments on BOSSBase image dateset showed that our proposed steganalytic method achieved the state-of-the-art performance against various steganographic schemes under different payloads.