Linguistic Steganography Detection Based on Perplexity

Abstract

Linguistic steganography is a technique for hiding information into text carriers. Most of the previous work on linguistic steganography was focused on steganography and there were few researches on steganalysis. Research on attacking methods against linguistic steganography plays an important role in information security (IS) area. In this paper, a linguistic steganography detection algorithm based on perplexity is presented. An experiment to detect text segments generated by linguistic steganography system NICETEXT is carried out. The result of our experiment is pretty promising. The experimental accuracy of our method on classification of generated text and normal text exceeds 99% when text size is larger than 400 bytes. Even for sentences, the experimental accuracy exceeds 85%.