最小二乘支持向量机的自适应数字水印算法  

Adaptive digital watermarking algorithm based on least squares support vector machine

作　　者：付港 李鸣[1] 

机构地区：[1]南昌大学机电工程学院,江西南昌330031

出　　处：《武汉工程大学学报》2015年第6期67-72,共6页Journal of Wuhan Institute of Technology

摘　　要：为了提高数字水印的安全性,提出了一种新的基于对比度掩蔽特性的数字水印算法.根据人眼视觉系统(HVS)的视觉特性,人眼对于亮度信号(Y)的改变较色度信号(U、V)的改变更加敏感,在图像学国际标准CCIR601当中,图像YUV空间中的Y∶U∶V通常会取4∶2∶2.首先引入对比度敏感函数模型,选择一些对比度函数值最大的子块,并升序排列;接着,对于前一半子块,使用最小二乘支持向量机建立图像邻域像素之间的关系模型;最后,通过动态地调整模型的输出值与中心像素值之间的大小关系,在另一半子块中实现水印的嵌入与提取.同时,还利用粒子群算法优化了最小二乘支持向量机的超参数.仿真实验表明,该数字水印算法具有着良好的不可感知性与鲁棒性,对一些常规的攻击操作也有着很强的抵抗能力.To improve the security of digital watermark, a novel digital watermarking algorithm based on the contrast masking properties was proposed. The human eye to the change of luminance signal （Y）is more sen- sitive compared with that of the chrominance signals （U,V）, according to the visual characteristics of human eye vision system （HVS）. In the YUV image space, the values of Y：U：V are 4：2：2 by the international image standard CCIR601. First, we introduced the contrast sensitivity functions, and chose some sub-images of the maximum contrast sensitivity functions, then ascended them. Second, for the front half of sub-images, the relational model of image neighborhoods pixels was established by the least squares support vector machine （LS-SVM）; Finally, the magnitude relation between the output and the central pixel of the model was dy- namically adjusted, meanwhile, the watermark was embedded and extracted in the other half of sub-images. The hyperparameters of LS-SVM were also optimized by the particle swarm algorithm. The digital watermark- ing algorithm has good robustness and imperceptibility, and has a strong resistance to aggressive operation.

关 键 词：数字水印 最小二乘支持向量机 不可感知性 鲁棒性 粒子群算法 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]