JPEG steganalysis based on bidirectional Markov model
Zhu Xiuming · Computer Engineering and Applications Journal · 2007
This paper presents a novel steganalysis scheme to attack JPEG steganography.The 600 dimensional feature vectors sensitive to data embedding process are derived from bidirectional Markov models in the DCT domain.The threshold Bayesian distance classifier is used to classify steganography in the high-dimensional feature vector space.In addition,SVM is used as a contrast.The experimental results have demonstrated that the proposed scheme outperforms the existing steganalysis technique in attacking modern JPEG steganographic schemes—F5,Outguess,MB1 and MB2.