An Anti-attack Watermarking Based on Synonym Substitution for Chinese Text

Current research of natural language steganographic algorithms based on synonymy substitution mostly focused on invisibility, but ignored robustness. However, automatic disambiguation of Chinese word senses (WSD) achieves high accuracy, the adversary could destroyed the watermark easily if he disambiguated the stego-text and did the synonym substitution again. In this paper, against the high accuracy of WSD algorithm, two indicators are proposed, which are Lexical similarity and sense similarity. For reducing the accuracy rate of WSD, we argued that it should choose the word, which is low Lexical similarity and high senses similarity, in the synonym substitution. Therefore, the automatically attack will failure. The experiment showed that the algorithm reduced the accuracy of WSD from 90.4% to 74.5%. The robustness of watermarking has been improved.