Abstract: A color image watermarking algorithm is presented which is in non-subsampled contourlet transform and singular value decomposition（NSCT-SVD）, by using scale-invariant features transform（SIFT） keypoints for image geometric correction. The algorithm uses the relations of the largest singular values of the NSCT low-frequency subblocks between gray and blue components, and embeds the watermark into the blue component, which has been encrypted by chaos system. During the watermarking detection, using the stable SIFT keypoints of the NSCT low-frequency in red component to correct the geometric attacks in the watermarked image, the watermark synchronous information is recovered and the watermark is extracted. Experimental results show that the proposed scheme is robust to a wide range of image processing operations such as adding noise, filtering and compression and all kinds of geometric attacks.

Key words: digital watermarking, Non-Subsampled Contourlet Transform（NSCT）, Singular Value Decomposition（SVD）, Scale-Invariant Features Transform（SIFT）, robustness

摘要： 提出一种利用尺度不变特征变换（SIFT）关键点对图像几何校正的非抽样Contourlet变换-奇异值分解域（NSCT-SVD）彩色图像水印算法。该算法利用蓝色与绿色分量的NSCT域低频子块的最大奇异值的关系，在蓝色分量上嵌入经过混沌加密的水印信息。水印检测时，先利用红色分量NSCT域低频系数上匹配的SIFT关键点信息对被检测图像进行几何攻击校正，恢复了水印的同步信息后再提取水印。实验结果表明，该算法对于噪声、滤波、压缩以及各类几何攻击具有较好的鲁棒性。

关键词: 数字水印, 非抽样Contourlet变换（NSCT）, 奇异值分解域（SVD）, 尺度不变特征变换（SIFT）, 鲁棒性