基于神经网络的小波域数字水印嵌入算法研究

提出一种结合神经网络将二值水印嵌入离散小波变换后的宿主图像中的新方法.为使算法具有更好的不可感知性和鲁棒性,进一步提高它的实用性,结合神经网络理论,创新地提出在小波域实现对数字水印嵌入.该方法是对宿主图像做离散小波分解,取分解后的近似分量作为嵌入位置.在其中随机的选取一些像素点及其邻域,利用神经网络对其进行建模及训练,通过修改其像素值嵌入水印信息.在嵌入之前对二值水印进行了Arnold变换来加密.实验结果表明,算法具有很强的抗几何攻击和承受其他图像处理操作的能力,不可感知性好,鲁棒性明显优于一般小波域嵌入算法,对数字水印的实现具有很强的参考价值.

An Algorithm of Digital Watermark Based on Neural Network in Wavelet Transform Domain

A new method of embedding a binary digital watermark into a discrete wavelet transformed image combining neural network is proposed.To get better imperceptibility and robustness,and to improve practicability,a digital watermark is embedded in wavelet domain with the new neural network theory in the proposed algorithm innovatively.The approximate part,which may be required from discrete wavelet transform of the original image,is taken as embedding location.Then the model is established and trained using neural network in pixels and their vicinities which are selected randomly from the approximate part.After that,watermark information is embedded by adjusting the value of pixels.Binary digital watermark is pre-treated by Arnold for encryption.The experiment results show that the proposed method has very strong ability of sustaining geometry attack and common image processing operations and has good imperceptibility.Its robustness is superior to other embedding algorithms in wavelet domain.It possesses very good reference value for the realization of digital watermark.