基于SVM与特征提取的鲁棒性数字水印算法

Robust Watermarking Scheme Based on SVM and Feature Extraction

为抵御大部分图像处理攻击,特别是几何攻击,提出一种改进的基于SVM与特征提取的鲁棒性数字水印算法.根据图像中的邻域像素之间的相关性,引入新的图像描述子和纹理描述子来提取图像特征.对样本点及其邻域像素的所有纹理变化进行训练,由邻域像素的和与方差组成训练集,训练好的SVM被用来分类一系列测试集,按照分类结果嵌入与提取数字水印.本文算法仿真试验结果表明,改进后新算法不仅具有较好的透明性,而且对中值滤波、叠加噪声等一般性处理和旋转、缩放、剪切等几何性攻击均具有更好的鲁棒性,训练时间和算法复杂度方面优于原有算法,提取的水印精度更高.

An image watermarking scheme is developed based on SVM（support vector machine） and feature extraction,which is robust against a variety of common image processing attacks and especially desynchronization attacks.A set of training patterns are constructed by employing two descriptors representing image features,according to the relationship of selected points to pixels neighborhood.The watermark embedding and extraction issues can be treated as a classification problem,and the binary results can be utilized to embed and detect watermark,which could tell the image texture variations.The embedded watermark is invisible to human eyes and receives better robustness under common image processing operations and some geometric distortions such as median filtering,noising or RST.The results show the scheme is of higher preciseness and lower complexity.