Steganalysis on Character Substitution Using Support Vector Machine

A new steganalysis method is proposed to detect the exists of hidden information using character substitution in texts. This is done by utilizing Support Vector Machine (SVM) as a classifier to classify the characteristic vector input into SVM. The most important step of this detection algorithm is the construction of a proper characteristic vector. Under the prerequisite that the secret bits to be embedded are uniformly distributed, the distribution of the characters used for hiding data has altered after steganographic process, thus the ratio of abnormal characters to normal characters is different in cover texts and stego texts. Experimental results demonstrate that this method can reliably detect whether there is hidden information in texts. The detection accuracy in experiment reaches as high as 96% while the embedding data rate is merely 20%.