摘要
提出了一种结合奇异值分解和神经网络的公开水印算法。首先对原始图像进行分块奇异值分解(SVD),然后建立最大奇异值系数与其他奇异值系数量化值之间的神经网络关系模型,最后通过调整最大奇异值与模型输出值之间的大小关系来嵌入水印信息。实验结果表明算法具有很好的水印透明性,对常见攻击(如JPEG压缩、平滑、加噪声和重采样等)具有较强的鲁棒性。
This paper presents a public watermarking algorithm combining single value decomposition （SVD） with neural network. We firstly use block single value decomposition to every block of original image, and then establish the relational model of neural network between the biggest single value coefficient and other coefficients. Finally a bit of the watermark is embedded by adjusting the polarity between the biggest value coefficient and the output value of the model. Experiment results show that this algorithm has good transparence of the embedded watermarking and is robust to attacks such as JPEG compression, smoothing, noise adding, and sampling.
出处
《微电子学与计算机》
CSCD
北大核心
2008年第2期1-4,共4页
Microelectronics & Computer
基金
湖南省自然基金项目(06JJ5098)
中国博士后基金项目(20060390882)
关键词
数字水印
神经网络
奇异值分解
公开水印
digital watermarking
neural network
singular value decomposition
public watermarking
