Research On Text Steganography Based On Chinese Character Stroke Fine-Tuning

Text steganography is an information hiding technique that embeds additional content into text,and is widely used in the fields of copyright protection and text steganography traceability.At present,there are three main directions of research on text steganography,namely,steganography for text images,text formats and text contents.Text Steganography Based on Chinese Character Stroke Fine-Tuning is a steganography method for text images,which uses the small differences of Chinese character strokes to embed secret information,which not only has good steganography and robustness,but also can maintain the readability and semantic integrity of text,so it has become one of the research hotspots in the field of information hiding.However,the method still has some problems,for example,the low accuracy of the word shape recognition method leads to the low efficiency and success rate of steganography extraction.Also,in the steganographic embedding process,there is a large optimization space for the existing Chinese character grouping algorithm.This dissertation focuses on two aspects of perturbation glyph recognition and Chinese character grouping algorithms,which include:(1)In this dissertation,a lightweight neural network-based Chinese character glyph recognition method is proposed for optimizing the extraction process of Chinese character strokes with fine-tuning steganography.First,a large amount of Chinese character glyph data is generated by manually fine-tuning the glyphs to construct a Chinese character glyph perturbation dataset.Then,a lightweight neural network is used to fit the glyph data to transform the glyph recognition task into an image classification problem.In addition,the method uses a fine-tuning strategy to quickly update the model parameters to accommodate the constant updates of the glyph files.Experiments demonstrate that the former method based on the MobileNetV2 model is more than 20% more accurate than existing Chinese character glyph recognition methods,and also its prediction efficiency is higher.(2)In this dissertation,we propose a grouping algorithm based on the independent probability of Chinese characters for optimizing the embedding process based on the fine-tuning steganography of Chinese strokes.The algorithm ranks Chinese characters in terms of word frequencies with the aim of making the sum of word frequencies in each group close and excluding the influence of words,which in turn makes the frequency of occurrence of Chinese characters close to the independent probability of the characters themselves.Experiments show that this grouping scheme can effectively improve the embedding rate of the steganography method when the number of text words is small.Compared with the existing grouping scheme,the embedding success rate of the method is also improved by more than 10%.