基于特征点模板的Contourlet域抗几何攻击水印算法研究

A Contourlet-Domain Watermarking Algorithm Against Geometric Attacks Based on Feature Template

抵抗几何攻击的数字水印是目前水印算法研究的热点与难点之一,一个微弱不可察觉的几何攻击就可能使绝大多数水印算法失效.文中对图像的Contourlet变换特性进行了分析和讨论,在此基础上提出了一种以特征点为模板的基于Contourlet的抗几何攻击水印算法,包括:(1)在Contourlet变换域最高阶方向子带的相同带内坐标中搜索纹理最丰富的位置作为嵌入位置,这样水印在被嵌入到图像重要区域的同时也保证其被嵌到纹理丰富的区域中;(2)根据嵌入点在低频对应位置的能量和高尺度同方向子带对应位置的纹理信息,确定了一种自适应的水印嵌入强度的策略,使得水印在嵌入位置和嵌入强度的选择上都保证了嵌入的水印具有很好的透明性和极大的鲁棒性;(3)改进了Harris-Laplace算子,利用该算子从含有水印的图像中提取出具有几何形变鲁棒性的特征点作为模板;(4)水印检测时,先利用特征点模板对应点之间的变换关系实现水印重同步,然后通过统计图像的相关性来验证水印的存在.仿真实验表明,所提出的水印算法不仅具有很好的透明性,而且对常见的图像处理和几何攻击具有很好的鲁棒性.

Image watermarking resistant to geometric attacks is the hotspot and challenging point in the state-of-the-art research on watermarking. Even an invisible affine transformation can invalidate most existing watermarking algorithms. This paper analyses and discusses the characteristics of Contourlet transform, and on the basis above, proposes a novel Contourlet-domain watermarking algorithm against geometric attacks based on the feature template matching. The proposed algorithm includes four important techniques： （1） Puting coefficients into one group with the same coordinate in each finest directional subhand. Then select the coefficient with the richest texture from each group as the coefficient to be embedded watermarking. Therefore, watermarking can be embedded into significant region as well as highly textured region of the host image; （2） For each coefficient of the selected subband, embeding watermarking with different strength adaptively adjusted in terms of the energy of its parent coefficient in the lowest subband, and the texture of its neighborhood coefficients in finer scale. The choice of embedding position and the embedding strategy above effectively guarantee good transparency and maximum robustness of the proposed algorithm; （3） Improving the Harris-Laplace operator, extract the feature points robust to geometric deformation as template from the embedded image using improved Harris-Laplace operator; （4） When detection, using the transformation between corresponding points in feature template to realize resynchronization, and determining existence of watermarking by statistical correlation between host image and watermarking. Experimental results show the proposed algorithm has good transparency and is very robust to common image processing and geometric attacks.