Research On Distortion-free Chinese Text Watermarking

Digita l watermark ing techno logy is an important branch of informatio n hiding,which pla ys a more important role in information security such as the copyrightprotection of digita l multimedia products, integrity ver ificatio n and source tracking.Text docume nts are used freque ntly in internet and they can be copied much s imp ly,which lead to a series of proble ms of information security. Traditio na l text digita lwatermarking methods mainly focused on embedding watermark methods, whichembedded watermark information by changing the format informatio n of thedocument or equiva lent substitutio n of content in the text. But some modifications oftext would be caused by these methods. So it is necessary to propose somewatermarking methods whic h will not bring in any change to text.In this paper, we investigate Chinese text d igita l watermarking methods. In orderto solve the proble ms addressed above, we propose two Chinese text digita lwatermarking methods, na me ly Chinese text watermarking method based on encodingmapping, and a Chinese text zero-watermark algorithm based on merging features ofsentence. In the first method, we map the waterma rk information into the text content.Firstly, a text is split to words and then we use tongyic i cilin to class ify words to somesemantics sets. Some high frequenc y semantics sets are chosen to map to binary stringby Huffma n cod ing. In the step of Huffman coding, the line information instead offrequenc y is used as the weight of high sema ntics sets, which improves the robustnessof the proposed algorithm. The second method is to extract the important informationfro m text to construct a watermark whic h is registered to the third party to achievecopyright a uthe ntication. In this method, it is necessary to do some preprocessing fortext document suc h as sentence segmentatio n, word segmentatio n, and word sensetags and so on. Then we calc ulate the sente nces entropy by the frequency o f se manticcode, the releva nce of sentences based on tongyici cilin and obtain the number ofuseful words as the length of sentences. Line weighting function is used to calculatethe fina l we ight of sentences. Selecting the high we ight se ntences, especia lly thewords whic h are nouns and verbs, are used to construc t a watermark. In the detectingstep, the s imilarity is calc ulated between the watermark extracted fro m the disputedtext and the watermark from CA to identify the text.In this paper, two Chinese text watermarking methods are proposed by taking advanta ge of the Chinese natural language processing techniq ues. The experimenta lresults demo nstrate the effectiveness of the proposed methods.