Information Hiding Based on Typing Errors

Abstract

Steganography hides secret in carrier. The carrier can be an image, a soundtrack, a video, a text or some other. Text steganography is sometimes more difficulty than others because of the high density of information carried by text. However, we found that most of the online texts is of a kind that tends to contain many typing errors, known as typos, but humans are very good at disambiguation, so these typos often do not frustrate understanding and sometimes it is even difficult to be recognized. This phenomenon can be found both in English and Chinese. In this paper, we propose a text steganography method using carefully injected typos and guarantee the security of the secret and the readability of the texts. With the help of a natural language processing (NLP) model named BERT, we can extract the secret message without the original text. Different from those format-based steganography algorithms, our method can resist format adjustments, OCR re-inputs, etc. Inspired by the text steganography in English text, text steganography in Chinese text with a similar principle is practical. Furthermore, social media platforms always contain many kinds of media, so Cross-Media or even Cross-Social-Network information hiding is practical when combining multi steganography algorithms.