Effective Linguistic Steganography Detection

Abstract

Linguistic steganography is an art of concealing secret messages. More specifically, it takes advantage of the properties of natural language, such as the linguistic structure to hide messages. In this paper, an effective method for linguistic steganography detection is presented. In virtue of the concepts in area of information theory, the method uses an information entropy-like statistical variable of words in detected text segment together with its variance as two classification features. The support vector machine is used as classifier. The method was centered on detection for small size text segments estimated in the hundreds in words. Its achievement is simple and its execution is fast and relatively accurate. In our experiment of detecting the three different linguistic steganography methods: NICETEXT, TEXTO and Markov-chain-based, the accuracy exceeds 90%. As a result, our method can be used as a common pre-detection method followed by a more specific and accurate detection method.