利用角点检测算法提取与图像旋转、平移、缩放等操作无关的Harris角点,对角点响应值与象素值的关系进行分析,进而角点增强响应值,以保证打印扫描前后检测到的角点具有较高的一致性,对角点构造特征向量作为图像的零水印信息。另外,针对已有局部特征描述子在打印扫描前后特征点匹配率不稳定这一问题,提出新的匹配算法,首先根据局部特征描述子进行粗匹配,然后用聚类法进行精匹配保证特征点匹配率。在水印检测时,对匹配成功的特征点进行遍历,将图像的局部特征信息和位置信息生成一维向量,对该一维向量作相似性运算,即可判别待检测图像是否含有零水印信息。实验结果表明,该算法对打印扫描等攻击具有一定的鲁棒性。
This paper introduces a zero watermarking method robust to print-scan attack .A method to enhance Harris corner is proposed in order to maintain high consistency .In addition , a new feature match is provided , which overcome the vulnerability of the original feature descriptor in face with print-scan attack .In watermark extraction , traversing the matching feature points of success , generating on-dimensional vector by using the local feature information and the position message of the feature points , with the similarity detection ( SD) value can determine whether the image to be detected contains watermarks .Experiment results show that the method provides robust watermark information after a print-scan attack .