基于神经网络和多小波变换的数字水印算法_英文_.pdfVIP

J ou rna l of Sou theast U n iversity (Eng lish Ed ition ) 　V o l. 23, N o. 2, pp. 2 11 - 2 15 June 2007　ISSN 1003—7985 D ig ita l wa termark ing a lgor ithm ba sed on neura l network in m ultiwavelet doma in 1, 2 3 W ang Zhen fe i 　　S ong Sheng li 1 ( C o llege of C om p u ter S c ience and Techno logy, H uazhong U n iversity of Sc ience and Techno logy, W uhan 430074, C h ina) 2 ( C o llege of Info rm a tion Eng ineering, Zhengzhou U n iversity, Zhengzhou 45000 1, C h ina) 3 ( D ep artm en t of C on tro l Sc ience and Eng ineering, H uazhong U n iversity of Sc ience and Techno logy, W uhan 430074, C h ina) Abstract:A novel b lind d ig ital w aterm ark ing algorithm based on neu ral netw orks and m u ltiw avelet transform is p resen ted. The host im age is decom p osed th rough m u ltiw avelet transform. There are fou r subb locks in the LL level of the m u ltiw avelet dom ain and these subb locks have m any sim ilarities. W aterm ark b its are added to low frequency coefficien ts. B ecause of the learn ing and adap tive cap ab ilities of neu ral netw orks, the trained neu ral netw orks alm ost exactly recover the w aterm ark from the w aterm arked im age. Exp erim en tal resu lts dem onstrate that the new algorithm is robust against a variety of attacks, esp ec ially, the w aterm ark ex traction does not requ ire the orig i