Linguistic steganalysis using the features derived from synonym frequency

Lingyun Xiang, Xingming Sun, Gang Luo, Bin Xia

Abstract: A linguistic steganalysis method is proposed to detect synonym substitution-based steganography, which embeds secret message into a text by substituting words with their synonyms. First, attribute pair of a synonym is introduced to represent its position in an ordered synonym set sorting in descending frequency order and the number of its synonyms. As a result of synonym substitutions, the number of high frequency attribute pairs may be reduced while the number of low frequency attribute pairs would be increased. By theoretically analyzing the changes of the statistical characteristics of attribute pairs caused by SS steganography, a feature vector based on the difference of the relative frequencies of different attribute pairs is utilized to detect the secret message. Finally, the impact on the extracted feature vector caused by synonym coding strategies is analyzed. Experimental results demonstrate that the proposed linguistic steganalysis method can achieve better detection performance than previous methods.