Research On Text Certification And Recoverable Watermarking Algorithm

Digital watermarking technology is a kind of technology which hides the specific producedand invisible marker information to digital products, such as digital texts, images and videos, so thatdigital products can be determined to the copyright owner and the integrity of contents. In recentyears, with the widespread application of electronic documents, researchers pay more attention totext watermark. In this thesis, taking advantage of the structural information of Chinese charactersand associative memory function of Hopfield neural network, it discusses a zero-watermarkalgorithm based on five-stroke splitting rules and a recoverable text watermarking algorithm basedon synonym substitution technology. The main work is as follows:(1) Discuss systematically the basic principles of text watermarking algorithm, the researchstatus and the advantages and disadvantages of various types of text watermarking technology.(2) Utilize five-stroke splitting rules of characters and propose a zero-watermark algorithmbased on the five-stroke splitting rules. In the algorithm, the Chinese text is turned into thatcomposed of letters according to the five-stroke coding information of Chinese characters, and thetotal number of characters and the largest occurrences of the letters between adjacent vowels isextracted to get feature information of the original text. Then the watermark is embedded by theway of zero-watermark. Simulation results show that the algorithm not only has better visibility, butalso can detect the type of tampering and locate tampering.(3) Utilize associative memory of Hopfield neural network and propose a recoverable naturallanguage text watermarking algorithm based on Hopfield neural network. The watermark isembedded by the way of synonym replacement, and synonyms which can’t cause the semanticambiguity after being replaced are selected by the way of calculating context collocation degree.The synonym thesaurus of a replaceable synonym is composed with the replaceable synonym in theoriginal text and all its seleted synonyms, and all the words in it are encoded according to theposition information. So the characteristic sequence of the original text is obtained from the codingof all the replaceable synonyms in the original text. Then the Hopfield neural network is trainedwith the characteristic sequence of the original text and the watermark in order to achieve therecovery of the original text. Simulation results show that the algorithm has good invisibility,robustness and recoverability. It not only can effectively resist all kinds of attacks against formats,but also can resist attacks against the replaceable synonyms, and better recover the original text.