Image steganalysis based on Markov model and feature fusion
JI Zhi-cheng · Kongzhi yu juece · 2009
An universal steganalysis scheme to attack JPEG steganography is presented.Difference arrays along horizontal,vertical and zigzag directions are formed by using the magnitudes of quantized block DCT coefficients.Markov process is applied to model these difference arrays so as to utilize the second order statistics for steganalysis.Feature vectors are derived from the Markov matrix to mining the quantified block DCT neighborhood coefficients' correlation.A weighted feature fusion method is used to fuse these feature vectors to improve the classification accuracy.The experimental results show that the proposed scheme has the advantage in detection rate in attacking OutGuess and F5.