Multi-class Markov models for JPEG steganalysis
Hao Zhang, Xijian Ping · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2013
Partially ordered Markov models based features were proposed in very recent years, which were shown to be quite effective in JPEG steganalysis. This paper presents an improvement of the original models. The proposed models here have two new characters. First, they are established on absolute values of coefficients instead of the values themselves. Second, the Markov models for coefficients were classified by comparing JPEG modes, not by directions. Besides, we recommended using Cartesian calibration technique to enhance the corresponding steganalytic features. Experimental results show that our proposed features outperform the original features, as well as some joint density features, in detecting several common steganographic algorithms.