摘要： 针对特征点发生偏差、水印检测效率降低的问题，提出一种将水印嵌入与特征点分开的算法。通过Harris特征点构造凸多边形区域，利用动态规划算法三角剖分凸多边形，水印信息嵌入在每个三角形的质心周围，少数且稳定的Harris特征点用来匹配原始图像与测试图像，估计计算几何变换参数，从而对测试图像进行校正。实验结果表明，该方法能改善水印对RST攻击的鲁棒性。

关键词: 动态规划, 凸多边形, 最优三角剖分

Abstract: Aiming at the inefficient watermark detection due to the variant feature points, a novel robust image watermark method is proposed based on dividing watermark insertion and the feature points. The Harris detector is utilized to extract feature points. According to the feature points, a convex polygon is adaptively constructed which is then optimally triangulated by dynamic programming. The watermark is embedded into the circles centering around the centeroid of each triangle. With the invariant feature points, the original image and the distorted images are matched so that the geometric manipulation can be computed. Experimental results show that the proposed method is considerably robust against Rotation, Scale, and Translation(RST) attacks.

Key words: dynamic programming, convex polygon, optimal triangulation

中图分类号: 

• TN911.73