Keyword-Based Robust Linguistic Steganography Against Paraphrasing Attacks

Current linguistic steganography techniques have achieved promising results in terms of resistance to steganalysis. However, they still face significant challenges in maintaining robustness against various forms of attacks, particularly paraphrasing attacks. To address this issue, this paper proposesKeyword-basedRobust LinguisticSteganography against paraphrasing attacks (KR-stega). First, to improve robustness, we leverage keywords as robust features of sentences while introducing locality-sensitive hashing (LSH) to partition the semantic space of words. By generating sentences conditioned on keywords, the method ensures accurate extraction of the secret message. Second, to improve imperceptibility, this paper introduces an imperceptible-enhancing sampling strategy. This strategy enhances the diversity of stegotext, thereby significantly lowering the risk of detection by steganalysis models. Experimental results demonstrate that this method not only enhances the robustness of stegotext against paraphrasing attacks, but also effectively improves the capability to resist steganalysis attacks.