Research On Reversible Text Watermarking Based On Natural Language

Recently, natural language text watermarking has been the research focus in text watermarking field. It features a strong robustness and a good security; but, to some degree, it changes the words or sentences of the text, which distorts the original meaning of the text and restricts its practical applications. In some critical application areas, such as military, legislation and literature fields, changes of the contents of the text are intolerable. This urgent practical demand requires the research on a watermarking scheme which can revert the contents of the text as well as embed watermark and extract watermark. At present, there is a reversible watermarking technology for image, which can restore the original image while extracting the watermarks. With reference to the technology for image, this thesis develops a study on reversible watermarking based on natural language and proposes two watermarking schemes which can not only embed secret information, but also restore the original contents of the text. The specific contents are as follows:First, by virtue of the thought of reversible watermarking for image, the thesis,for the first time, puts forward the concept of reversible text watermarking. And then, based on the exploitation of the differences between text and image in terms of embedding watermark, two reversible watermarking schemes based on natural language are proposed:The first reversible watermarking scheme is based on an invertible integer transform. The invertible transform used in the algorithm has the following feature:after applying the inverse transform, each element of the integer pair is one smaller than that of the original integer pair, if the transformed pair through the forward transform losses the last bit. Thus,the last bit of an integer pair can be used to embed one bit of watermark. The algorithm firstly converts the transformable words or sentences into the form of integers and splits the integers gotten from the text into pairs, then embeds the watermark into the converted integer pairs or extracts the watermark, finally gets the text with watermark or recovers the original text using the updated integer pairs. The experiment results show that the algorithm can embed enough extra information to identify the copyright of the text after embedding the information used for restoring the original contents of the text.The second reversible watermarking scheme proposed here is based on an improved difference expansion method. The improved difference expansion method uses the last two bits, rather than the last bit, of the index of a synonym word or a changeable sentence. Meanwhile, if the integer pairs are unchangeable after the process of difference expansion, the algorithm applies the differences of integer pairs as the abscissa values of the histogram and shifts them, which can achieve a higher embedding capacity. Compared with the traditional difference expansion, this algorithm improves the embedding capacity. Thus it is applicable to text reversible watermarking.