设计了基于可逆Walsh-Hadamard变换与预测误差扩展的图像水印算法。将载体图像分割为4×４的非重叠子块, 引入Walsh-Hadamard变换, 对每个子块进行处理, 得到其域Hadamard, 提取其直流分量与交流分量系数;基于Adaline神经网络, 确定一个准确的预测系数, 构建线性预测函数, 计算每个交流系数的误差;对交流系数误差完成二进制表示, 根据水印信息对误差进行扩展;考虑像素的溢出或下溢问题, 从所有子块中指定一个平滑扩展块, 通过设计水印嵌入方法, 将水印信息嵌入到扩展块中, 再借助Walsh-Hadamard逆变换, 输出水印图像;构建水印提取机制, 从水印图像中提取初始水印信息。实验结果显示: 与当前水印图像相比, 所提算法具有更高的不可感知性与更低的失真度。

The new image watermarking algorithm based on reversible walsh-Hadamard transform and prediction error extension was proposed. The carrier image was segmented into non-overlappingsub-blocks, and the Hadamard domain was got by introducing the Walsh-Hadamard transform to process each sub-block for extracting DC component and AC component coefficient. An accurate prediction coefficient was determined based on the prediction error expansion and adaline neural network to construct linear predictive function to predict AC coefficients and calculate the error of each AC coefficient. The binary representation of AC coefficient error was finished, and the error expansion was done based on watermark information. A smooth extension block from all sub blocks was specified by considering the problem of pixels overflow or underflow, and the watermark information was embed into an extension block by designing the watermark embedding method so that watermark image was outputted by reversible walsh-Hadamard transform. The watermark extraction mechanism was constructed to extract watermark information from the watermark image. The experimental results show that this algorithm had higher perception and lower distortion compared with the current watermark image.