基于SIFT特征点匹配校正的抗几何攻击水印算法

摘要

Digital products with watermarking may suffer from a variety of attacks in the transmission process.Among them,the geometric attack could destroy the synchronization between the watermark and the host image and so to fail the watermarking algorithm.Therefore,how to design watermarking algorithms that could avoid geometric attacks have been the hotspot of recent research for digital watermarking technologies.In this paper,a geometric correction method is proposed which is based on rotation,scaling and translation invariance of the scale-invariant feature transform (SIFT) feature points.Geometric correction of watermarked images is realized by scale features and coordinate relations of SIFT feature points.Contourlet transform is applied to the host image and the low-frequency subband is divided into blocks when the scrambled watermarking information is embedded.Each block is then decomposed by SVD and all singular values are modified by quantifying the Euclidean norms of the singular value vectors of each sub-block.When watermarks are extracted,the attacked information will be synchronized by geometric correction.Simulation results show that the algorithm with fine invisibility has a larger watermark capacity with good robustness to resist conventional attacks and geometric attacks.%嵌入水印的数字产品在传播过程中可能遭受到各种攻击,其中几何攻击破坏了水印和图像空间坐标的同步性,很容易使得水印算法失效.因此,抗几何攻击的水印算法成为了数字水印技术研究的热点和难点.本文利用图像尺度不变特征变换(SIFT)特征点所具有的旋转、缩放和平移不变性,提出了一种几何校正方法,该方法根据SIFT特征点的尺度特征和坐标关系实现带水印图像的几何校正.水印嵌入时首先对宿主图像Contourlet变换的低频子带进行分块奇异值分解,然后通过对各块奇异值向量的欧式范数的量化,改变所有奇异值实现水印信息的嵌入.水印提取时先对图像进行相应的几何校正,使得水印信息在受攻击后也能得到同步,从而提高了水印图像的抗攻击能力.仿真结果表明,该算法不可见性良好,水印容量较大,而且对于常规攻击和多种几何攻击都具有较好的鲁棒性.