Steganalysis Based on Image Contents
Xijian Ping · 2013
Digital images are important covers in information hiding.It has been shown that steganalytic methods perform differently on images with different contents.Based on regionally stationary Markov characteristics of natural images,a new steganalysis method based on image segmentation is proposed.Quad-tree segmentation based on local Variance is used to locate areas sensitive to message embedding.Features based on run length histogram are then extracted and optimized.Finally,detection of steganography is done using support vector machine.Experimental results show that the proposed steganalysis method outperforms previous methods in detection accuracy.