摘要
为了进一步提高基于支持向量机(SVM)水印算法的鲁棒性,提出了一种基于复Contourlet域的SVM和Krawtchouk矩的双水印算法。首先从RGB宿主图像中提取B分量和G分量,并且充分利用Krawtchouk矩不变量的平移、旋转、缩放不变性和Krawtchouk矩良好的局部重构特性,计算B分量图像的Krawtchouk低阶矩不变量,由此构造鲁棒水印;然后对G分量图像进行两级复Contourlet分解,在其低频分量中,利用SVM建立图像尺度内的局部相关性训练模型,并根据预测结果自适应地实现数字水印图像的嵌入和提取。大量实验结果表明,本文算法不仅具有较好的不可感知性,而且对中值滤波、加性噪声和JPEG压缩之类的常规图像处理,以及缩放、旋转和剪切等几何攻击,均具有较好的鲁棒性能,其性能优于基于小波域的SVM和基于Contourlet域的SVM水印算法。
In order to further improve the robustness of watermarking algorithm based on support vector machine （SVM）, a double watermarking algorithm based on SVM in complex Contourlet domain and Krawtchouk moment is proposed in this paper. Firstly, the blue component and green component are ex- tracted from the RGB host image. The algorithm makes full use of the invarianee of Krawtchouk moment invariants to translation, rotation and sealing, and the better local reconstruction characteristics of Kraw- tchouk moment. The lower order Krawtchouk moment invariants of the blue component are calculated for construction of robust watermarking. Then the green component image is decomposed by two-level complex Contourlet transform. In the low-frequency components, the local correlation training model of image in the same scale is established by using support vector machine. The watermarking image is adap- tively embedded and extracted according to the prediction results of the established model. A large num- ber of experimental results show that the proposed algorithm is not only invisible but also robust against the common image processing, such as median filtering, Gaussian noise and JPEG compression, and a- gainst some kinds of geometric attacks,including watermarking algorithm based on support vector rotation, scaling and clipping. It outperforms the image machine in wavelet domain or Contourlet domain.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2014年第11期2170-2177,共8页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(60872065)
江苏省社会安全图像与视频理解重点实验室(南京理工大学)开放基金(JSKL201302)
江苏高校优势学科建设工程资助项目
