基于支持向量机的鲁棒盲水印算法

Robust blind watermarking scheme on support vector machine

提出了一种基于支持向量机的鲁棒盲水印算法。该算法首先用多尺度Harris-Laplace检测算子从载体图像中提取出稳定的特征点,然后根据特征自适应确定局部特征区域,在特征区域选择一些点作为嵌入水印的点,结合图像的邻域相关性,根据灰度图像特点,选取特征向量作为SVR训练模型,进而利用SVR进行预测,调节嵌入点的像素值进行水印的嵌入和提取。实验结果表明,用该技术嵌入水印后的图像具有很好的图像感知质量,对常规信号处理和去同步攻击特别是JPEG压缩具有较强的鲁棒性。

A new robust blind watermarking scheme based on SVM is proposed. Firstly, the multi-scale Harris-Laplace detector is utilized to extract feature points, and then the local characteristic regions are adaptively constructed according to the characteristic scale theory, selected some points from the regions as embedded points. The watermarking embedding and extraction are achieved by using a preparational technique based on support vector regression model to adjust the values of the selected pixels. The model is trained by learning the relationship between the selected pixel and statistical feathers of neighboring area. Experimental results show that the presented scheme has good image perceptual quality and high watermark robustness to common image processing operation and desynchronisation attacks, specially, to JPEG compression.