一种使用机器学习方法的数字水印算法

针对目前数字水印算法存在的不足,本文将离散小波变换和奇异值分解相结合,提出了一种基于机器学习的图像数字水印算法.首先将载体图像进行一级小波变换,提取其低频子带图像对其进行4×4分块处理,然后对每一分块进行奇异值分解后嵌入水印,并提取特征向量用于最小二乘支持向量机的训练,训练好的最小二乘支持向量机用于自适应最大水印嵌入强度的计算以及水印的盲提取.实验选取三张512×512的标准测试图像以及64×64的二值水印图像对算法的透明性与鲁棒性进行测试.实验结果证明,图像具有很好的透明性,PSNR达到了63.71dB,针对旋转、剪切、JPEG压缩、高斯噪声等常规攻击手段时,算法能保持较强的鲁棒性.

Digital Watermarking Algorithm Using Machine Learning Method

In view of the shortcomings of current digital watermarking algorithms,this paper proposes a machine learning based image digital watermarking algorithm by combining discrete wavelet transform(DWT)and singular value decomposition(SVD).Firstly,the image is processed by DWT,and then the low-frequency sub-band image is extracted for 4×4 block processing.Then the watermark is embedded by SVD for each block,and the feature vector is extracted for the training of LS-SVM.Trained LS-SVM is used to calculate the adaptive maximum watermark embedding strength and blind watermark extraction.Three 512×512 standard test images and the 64×64 binary watermark image are selected to simulate the transparency and robustness of the algorithm.The experimental results show that the transparency of the images is very good,and the PSNR is up to 63.71dB.The algorithm can maintain strong robustness against conventional attacks such as rotation,cropping,JPEG compression,Gaussian noise and so on.