Statistical Features for Universal Steganalysis on Color Images

In this paper we propose three types of features which can be used to realize efficient universal steganalysis on color images. After measuring the deviation of DCT coefficient distribution,the correlation of smoothness in spatial domain,and the correlation between different components,a 10-dimentional feature vector is produced for each color image. Those features are sensitive to steganographic process and the feature vector,having a low dimension,is easy to calculate. Then,a SVM classifier is built to distinguish the stego-images from the original images. Experiments show that the universal steganalytic approach can effectively detect the presence of secret message hidden by Jsteg,F5 or MB steganography,and can identify the steganographic techniques used.