基于SIFT特征区域的支持向量机数字水印方法  

Digital watermarking scheme based on SVM of SIFT features

作　　者：房小兆 张巍[1] 滕少华[1] 韩娜[1] 

机构地区：[1]广东工业大学计算机学院,广东广州510006

出　　处：《计算机工程与设计》2011年第11期3609-3612,共4页Computer Engineering and Design

基　　金：广东省自然科学基金项目(06021484;9151009001000007;9451009001002777);广东省科技计划基金项目(2008A060201011);广州市越秀区科技计划基金项目(2007-GX-075)

摘　　要：针对传统水印算法抵抗几何攻击和组合攻击弱的缺点,提出基于SIFT特征区域的支持向量机数字水印方法。根据SIFT提取的特征点选取合适的特征区域,利用回归型SVM较好地建立特征区域内邻域像素之间的关系模型,并通过调整模型的输出值与中心像素值之间的大小关系来自适应地嵌入或提取水印。实验结果表明,该方法对一般几何攻击和组合攻击有较强的抵抗能力。For the problem ofweak against geometric attacks and portfolio attacks of traditional watermarking algorithm, a digital watermarking scheme based on SVM of SIFT features is proposed. Scale invariant feature transform （SIFT） is first adopted to extract feature points and to construct appropriate local circular feature regions. And it can learn the relationship between the selected pixel and its neigh- boring pixels well with support vector regression. Then, the watermark is embedded or extracted by adjusting the value between the se- lected pixel and the actual output of the trained SVR. Experimental results show that the proposed scheme is robust against geometric attacks and portfolio attacks.

关 键 词：数字水印 特征区域 支持向量机 鲁棒性 关系模型 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]