Research On Coverless Text Information Hiding Technology Based On Specific Word Sets

The problem of data security becomes more and more important with the development of the Internet of things.Information hiding technology,which is one of the methods of covert communication,has more and more demand making its development face new challenges.The information hiding technology based on text develops slowly due to the lack of sufficient redundancy.Compared with the modified text information hiding algorithm,the coverless information hiding algorithm has higher embedding capacity and can resist steganalysis attack.However,the coverless information hiding algorithm still has some limitations.Thus,a steganography method based on arithmetic coding and large-scale neural language model is proposed.The specific work is as follows:In order to deal with the low imperceptibility of the steganography text generated by the current steganography methods and the large difference in the statistical distribution between the steganography text and the natural text,a steganography coding method based on dynamic word selection strategy is proposed to make the steganography text achieve good imperceptibility.In the specific design process,the large-scale pre-trained language model GPT-2 is used as the generation model of the steganography method.Combined with the conditional probability distribution output by the language model,the conditional probability distribution of words is truncated by Top-K sampling to form an initial candidate pool;According to the word probability distribution at different generation times,a dynamic word selection strategy is designed to determine which word can enter the new candidate pool by setting a threshold for the variance of the word probability distribution in the initial candidate pool.At the same time,an imperceptible coefficient is introduced to guarantee imperceptibility.The arithmetic coding method with better imperceptibility is selected as the coding algorithm of the proposed algorithm because of the imperceptibility of huffman coding and arithmetic coding.Therefore,steganography text is generated.Experimental results show the embedding rate,text quality and statistical distribution under different variance thresholds and imperceptibility coefficients.Comparative experiments with the other four selected baseline models show that the proposed algorithm has better performance in ensuring imperceptibility,the algorithm can reduce confusion and improve text quality,but has less sacrifice in hiding capacity.