Abstract: In order to improve the robustness of traditional digital image watermarking based on Singular Value Decomposition（SVD）, this paper proposes a watermarking scheme resisting to rotation attack based on Radon transform and SVD in wavelet domain. Wavelet transform is applied to the host image, then decomposes low frequency component of host image. It embeds the binary watermark image after affine transform scrambling into the singular value of the image. It uses odd-even quantization algorithm to embed watermark, which achieves embedding binary image blindly. Meanwhile before extracting watermark, it uses Radon transform detection algorithm to correct the text image, then watermark information is extracted. The experimental results demonstrate the proposed method is better than traditional algorithms based on SVD in commonplace signal processing, such as adding noise, filtering and JPEG compression etc. It is also robust enough for rotation attacks.

Key words: singular value decomposition transform, geometric attacks, Radon transform, robustness

摘要： 为了提高传统基于奇异值变换（SVD）的数字水印抗几何攻击能力，提出一种在小波变换域将Radon变换和奇异值变换相结合的抗旋转攻击鲁棒性水印算法。将宿主图像进行小波变换，对变换后的低频子带进行奇异值分解，将经过仿射变换置乱后的二值水印图像嵌入到奇异值中。在水印嵌入操作上采用了奇偶量化嵌入算法从而实现了二值水印图像在水印检测时的盲提取；同时在水印检测之前，利用Radon变换检测算法对待检测图像进行几何校正，然后提取水印信息。实验结果表明，该算法对于噪声感染、滤波、JPEG压缩等常规信号处理的鲁棒性优于传统的基于SVD的数字水印算法，同时对于旋转几何变换具有很好的鲁棒性。

关键词: 奇异值变换, 几何攻击, Radon变换, 鲁棒性