一种基于神经网络和小波变换的盲水印算法

A Blind Watermarking Algorithm Based on Neural Network and Wavelet Transformation

提出一种新的运用神经网络将水印嵌入离散小波变换后的宿主图像中的数字水印算法。算法创新地提出选择水印嵌入点周围的8个邻域像素点作为神经网络的样本输入,这样可以更好地增强水印的透明性。该算法首先对宿主图像做离散小波变换,取低频子带作为嵌入位置,运用密钥随机选取水印的具体嵌入坐标,提取其邻域的像素点,利用建模好的神经网络对其进行训练,通过修改其像素值嵌入水印。提取时利用确定的神经网络实现水印的盲提取。水印嵌入之前运用Arnold变换进行了置乱处理。实验结果表明,该算法具有较好的不可见性和鲁棒性。

A new algorithm of embedding a digital watermark into a discrete wavelet transformed image using neural network. Eight neighborhood pixels are innovatively chosen as the input of neural network in the algorithm, which can better improve the transparency of watermark. The low frequency suband images,which can be required from discrete wavelet transform of the original image,is taken as embedding location. Randomly chooses the watermark＇s concrete embedded coordinates with the key, and extract their neighborhood points. Then they are trained by the modeled neural network. The watermark is embedded by adjusting the value of pixels. While the model of neural network is made certain,the blind watermark is extracted. The watermark is confused by Arnold with the key before being embedded into the host image. The experiment results show that this algorithm is imperceptible and robustness.