基于频域变换与几何失真校正的图像水印算法

为改善水印系统的稳健性,提出了基于多元频域变换的彩色图像水印方案.首先,引入仿射映射来混淆水印数据;再确定载体的最大中心区域,利用四元离散Fourier变换对其处理,获取1个实部系数矩阵与3个虚部系数矩阵;考虑彩色载体的亮度、纹理与边缘掩码,计算数据融入因子;并以此建立同步水印隐藏方法,把加密数据植入实部系数矩阵的低频系数中,形成水印目标;基于不同的内容变换,形成训练数据;并基于机制,对其完成二级分解,利用分布函数来拟合6个系数的经验直方图,最终将其形状与尺度参数视为水印系统的稳健特征矢量.借助这些稳健特征来训练支持向量机,从而得到变换参数值,以校正受攻击的水印图像;最后,计算校正水印图像的最大中心区域,联合四元离散Fourier变换,设计水印提取方法,准确复原二值水印.测试数据表明:与当前图像水印技术相比,面对几何变换攻击,所提算法具有更高的不可感知性与鲁棒性,其输出的水印图像对应的差分图较为理想,复原水印失真度最小.

On Image Watermarking Algorithm Based on Frequency Domain Transformation and Geometric Distortion Correction

In order to enhance the ability of image watermarking to resist unknown attack, an color image watermarking algorithm based on multi-frequency domain transformation has been proposed in this paper. Firstly, affine mapping was introduced to encrypt the binary watermark information for destroying the pixel spatial relationship. Then the maximum center area of the carrier image was determined so that the real coefficient matrix and three imaginary coefficient matrices matrix were obtained based on quaternion discrete Fourier transform. The data integration factor was calculated based on the brightness, texture and edge mask of the image. Then a synchronous watermark embedding method was designed according to the integration factor for embeding the encrypted data into the low frequency coefficients in the amplitude information matrix for forming the watermark image. The training dates were constructed by appling different contents transformations. The two level decomposition of quaternion wavelet transform was done on them, and the BKF distribution function was used to fit the empirical histogram of the 6 DWT coefficients for considering the shape and the scale parameters as the robust feature eigenvector of the watermark image. With the help of these robust feature, the support vector machine was trained to get the transform parameter values for correcting attacted watermark images. Finally, the maximum center area of the watermark image was calculated, and the watermark extraction method was designed based on quaternion discrete Fourier transform to accurately restore the binary watermark. The test data show that this algorithm has higher uncertainty and robustness in the face of geometric transformation attacks which the differential image of the output watermark image was more ideal, as well as the distortion of the restored watermark was minimal compared with the current image watermarking technology.