一种基于奇异值分解和反传神经网络的多小波域数字水印算法  

Multiwavelet-based image watermarking using SVD and BP neural network

作　　者：李国良[1] 谭月辉[1] 玄克诚 李志勇[3] 彭涛[3] 

机构地区：[1]军械工程学院计算机工程系 [2]65301部队装备部 [3]军械工程学院训练部

出　　处：《河北工业科技》2010年第4期219-222,共4页Hebei Journal of Industrial Science and Technology

摘　　要：针对图像经过CARDBAL多小波变换后,各层系数的能量平均分配,且其同一尺度、同一方向上的4个子块系数相近这一特性,结合奇异值分解和BP神经网络将二值图案水印嵌入到多小波系数的高频分量中。实验表明,该算法有良好的不可感知性和强健的鲁棒性。A blind watermarking algorithm based on CARDBAL multiwavelet transformation,SVD and BP neural network is proposed.In this scheme,the original image is firstly decomposed by multiwavelet transformation,then the relationship among singular values of subblocks in high frequency areas is learned by BP neural network.Finally a watermark signal is embedded into the multiwavelet domain partly by adjusting the relationship among singular values of these subblocks.The experimental results prove the robustness and imperceptibility of this method.

关 键 词：数字水印 多小波 奇异值分解 BP神经网络 

分 类 号：TN911.73[电子电信—通信与信息系统]