基于二代Bandelet变换与稳定特征检测的鲁棒图像水印算法

当前基于特征点的图像水印技术主要是将水印信息隐藏到采用固定值构建的特征不变区域中,其抵御几何变换能力不佳。对此,设计二代Bandelet变换耦合稳定特征检测的图像水印算法。引入尺度不变特征变换方法,提取载体图像中潜在的特征点;根据特征点的Hessian矩阵对应的特征值,构建不稳定特征点消除规则;根据特征点的尺度因子,计算圆形半径,建立局部特征不变区域;定义鲁棒特征区域选择机制,找出合适的水印嵌入位置;引入二代Bandelet变换,处理鲁棒特征区域,获取并选择出重要的Bandelet系数;设计水印嵌入方法,将经过仿射映射加密后的二值水印隐藏到重要的Bandelet系数,获取水印图像;检测水印图像的稳定特征点,联合二代Bandelet变换,从水印图像中复原二值水印。测试结果显示:与已有的鲁棒水印方案相比,该方案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更为理想,在多种几何攻击下的复原水印失真度最小。

ROBUST IMAGE WATERMARKING ALGORITHM BASED ON THE SECOND GENERATION BANDELET TRANSFORM AND STABLE FEATURE DETECTION

At present,image watermarking technology based on feature points mainly hides the watermarking information into the feature invariant region constructed by fixed values,which makes the ability of resisting geometric transformation poor. In order to solve the problem,we designed a robust image watermarking algorithm based on the two generation Bandelet transform coupled with stable feature detection. Scale invariant feature transform was introduced to detect all potential feature points in the image. According to the Hessian matrix corresponding to the feature points,the unstable feature point elimination rules were constructed. The circular radius was calculated according to the scale factor of the feature points,and the local feature invariant regions were established. A robust feature region selection rule was defined to determine the appropriate watermark embedding location. The second generation Bandelet transform was introduced to deal with the robust feature region,and the important Bandelet coefficients were obtained and selected.Then we designed the watermark embedding method to hide the binary watermark encrypted by affine map to the important Bandelet coefficient to get the watermark image. We detected the stable feature points of the watermarking image,combined the second generation Bandelet transform,and restored the binary watermarking from the watermarkingimage. The test results show that the proposed watermarking scheme has higher imperceptibility and robustness than the existing robust watermarking schemes. The difference image corresponding to the output watermarking image is more ideal,and the distortion of the restored watermarking under various geometric attacks is the smallest.