The existing watermarking technology for digital image has defects more or less due to some unavoidable reasons. There are some issues in various links,such as image sampling,watermark embedding and extracting,image synthesis,etc. SVM model is used in this paper for classifying corresponding categories by training a large number of image patches with differ⁃ent textures and brightness to realize variable intensity watermark embedding. Moreover,the genetic algorithm is improved fur⁃ther by retaining the some best individuals. Meanwhile,sampling modes are reformed to embed image watermarking in the DCT domain. The experiment results show that the method makes the image with embedded watermark has a higher PSNR value,and has preferable anti⁃attack ability against JPEG,Gaussian noise,rotation,low⁃pass filtering and histogram equalization.%由于一些不可避免的因素,现有的数字图像水印技术或多或少的存在各种缺陷,在图像采样、水印嵌入、图像合成、水印提取等各个环节都存在值得商榷的地方。采用支持向量机(SVM)模型,通过对大量不同纹理与亮度块的训练,使得图像块通过SVM得出相应的类别,从而实现水印强度的可变嵌入,并且,通过保留个别最佳个体进一步改进遗传算法,同时改变采样方式,在图像分块的DCT域中嵌入水印。实验证明,该方法使得嵌入水印图像与原始图像有较高的PSNR值,同时对JPEG、高斯噪声、旋转、低通滤波、直方图均衡化等具有较好的抗攻击能力。
展开▼