Research On Carrier-free Steganography Of Text Based On Language Features

Text steganography can be categorized into steganography by text modification and coverless steganography.Because the coding redundancy was very low in a text,even slight changes to it may be discovered by people or machines,so the steganography by text modification has a low embedding capacity and poor security.Coverless steganography can be divided into generation and detection methods.Steganography by text generation generated some texts to hide a secret by utilizing natural language processing technology.Nowadays,the natural language generation technology was relatively mature,but the generated long text still has poor readability,which was easy to be recognized.Steganogaphy by text selection was to find the text corresponding to the secret message based on the given rules,and the whole process does not modify or generate the text.There is a significant safety improvement.However,this type of method usually employed a single characteristic and has a limited scope of application.After a coverless retrieval method based on language features was proposed,a steganography framework was proposed to integrated various steganography methods using different language features to improve the embedding capacity.The specific works were as follows:(1)A coverless steganography method was proposed by employing the parity of the occurrence frequency of"middle level and oblique"in Song Ci.The mapping relationship between secret information and the carrier of Song Ci was established by encoding the frequencies of different rhythms in Song Ci.In order to hide the secret message,the binary secret message,which the song tune can represent,was determined by mapping relation.This method did not require a vast dictionary library,had good invisibility and robustness,and had an embedding capacity of 3 bits/song tune.(2)A general framework of text coverless steganography based on different language features was proposed to integrate various steganography features for steganography.Each steganographic feature divided the text into 2⁹)categories,represented by 2⁹)two levels of complete 2⁹)trees,each of which was called a meta-tree.Establish mapping relationships among secret messages,meta tree,and text.The meta-tree sequence was obtained by matching the carrier text with the binary string fragments of the secret message,and the steganographic tree was constructed by using the meta-tree sequence to realized covert transmission.The framework had high embedding capacity,strong invisibility,a high success rate of steganography,and a wide application range.The proposed steganography methods made full advantages of coverless steganography,including the perfect invisibility,robustness,and effective resistance to the steganography analysis,and achieved a hiding success rate of 100%.The proposed steganography framework could integrate multiple language steganography features,which improved the framework's applicability and provided a foundation for large-capacity fasted steganography,and more language steganography features could be explored on the framework.