基于SVD和神经网络的鲁棒水印算法

本文提出了一种基于奇异值分解和小波包分解相结合的全新水印算法。综合利用奇异值和方差的特征来对宿主图像进行预处理之后,提取出两个具有不同掩蔽效应的子图,分别对子图的奇异值和小波包系数使用抖动调制方法嵌入不同强度的水印。本文算法不仅实现了水印的盲提取,而且实现了水印鲁棒性与透明性的最佳折中。本文的另一创新之处是提出了一种基于神经网络的新的水印复原方法。实验结果表明本文方法较传统方法而言有更高的灵活性和鲁棒性,尤其具有很强的抗JPEG压缩能力。

A Robust Watermarking Algorithm Based on SVD and Neural Network

In this paper,we present a novel digital image watermarking algorithm based on singular value decomposi- tion and wavelet packet decomposition.A new method which successfully combines the characteristics of singular val- ues and variance is developed to preprocess the host image to obtain two sub-images with different masking effect.The application of dither modulation embedding scheme makes our method achieve the goal of watermark blind-extraction.Another significant innovation of our algorithm is that we propose a different means of watermark recovery based on neural network,which contributes to accurate data tracing.Simulation results show that our method is more flexible than traditional methods and fulfills a good compromise between the robustness and transparency.