Steganalysis of HMPD reversible data hiding scheme

Based on feature mining and pattern classification, this paper presents a steganalysis technique against reversible data hiding by histogram modification of pixel differences scheme (HMPD). HMPD reversible data hiding scheme involves the modification of pixel differences, which introduces artifacts into the pixel-difference histograms. It is found that four-way pixel difference features are sensitive for HMPD embedding algorithm. These features are used to design a specific steganalysis method for detecting HMPD reversible data hiding scheme. Support Vector Machine (SVM) classifiers are trained to discriminate stego-images from cover images and estimate the secret keys of a stego image. The experimental results show that the proposed scheme not only successfully attacks HMPD reversible data hiding scheme in various embedding levels, but also accurately estimates the secret keys.