NSCT-GBT-SVD结合特征区域的鲁棒水印算法  

NSCT-GBT-SVD Combined with Feature Region Robust Watermarking Algorithm

作　　者：张天骐[1] 马焜然 杨宗方 邹涵 ZHANG Tianqi;MA Kunran;YANG Zongfang;ZOU Han(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications(CQUPT),Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《信号处理》2023年第12期2241-2253,共13页Journal of Signal Processing

基　　金：国家自然科学基金项目(61671095,61702065,61701067,61771085);信号与信息处理重庆市市级重点实验室建设项目(CSTC2009CA2003);重庆市自然基金项目(cstc2021jcyj-msxmX0836);重庆市教育委员会科研项目(KJ1600427,KJ1600429)。

摘　　要：针对大多数水印算法抵抗几何攻击(旋转、平移、缩放和剪切)不佳,本文提出了一种非下采样轮廓波变换(Non-Subsampled Contourlet Transform,NSCT)-图变换(Graph-Based Transform,GBT)-奇异值分解(Singular Value Decomposition,SVD)结合(多尺度多方向结构张量-模糊C均值聚类)特征区域的鲁棒水印算法。首先利用NSCT变换得到低频区域,接着建立多尺度多方向结构张量寻找稳定的特征点,并用模糊C均值聚类将特征点划分为三簇,构造每个簇中稳定且不重叠的特征区域。然后利用图像局部归一化操作构造水印嵌入区域,将水印嵌入区域分块并对分块区域进行GBT得到稳定的系数矩阵。最后,利用SVD将水印信息相加嵌入到系数矩阵中。仿真实验表明,本文算法在峰值信噪比高于44 dB的情况下,对常规信号攻击和几何攻击均有很好的鲁棒性,特别是在面对旋转、平移、剪切、缩放以及两两组合的几何攻击时,提取的水印归一化相关系数均在0.94以上。In view of the poor resistance of most watermarking algorithms to geometry attacks(rotation,translation,scaling,and shearing),this paper proposed a Non-Subsampled Contourlet Transform(NSCT)-Graph-based transform(GBT)-Singular Value Decomposition(SVD)combined with(multi-scale multi-direction structure tensor-fuzzy C-means clustering)feature region robust watermarking algorithm.Firstly,NSCT transform was used to obtain the low-frequency region,and then the multi-scale and multi-direction structure tensor was established to find the stable feature points.The feature points were divided into three clusters by fuzzy C-means clustering,and the stable and non-overlapping feature regions in each cluster were constructed.Then the watermark embedding region was constructed by the local normalization operation of the image,and the watermark embedding region was divided into blocks and the stable coefficient matrix was obtained by GBT of the block region.Finally,SVD was used to embed the watermark information into the coefficient matrix.Simulation experiments show that the proposed algorithm has good robustness to both conventional signal attacks and geometric attacks when the peak signal-to-noise ratio is higher than 44 dB.Especially,when facing the geometric attacks of rotation,translation,cutting,scaling and pairwise combination,the extracted watermark normalized correlation coefficients are all above 0.94.

关 键 词：图像水印 特征区域 非下采样轮廓波变换 图变换 图像局部归一化 

分 类 号：TN911.7[电子电信—通信与信息系统]