基于不变质心及SIFT特征点的抗几何攻击零水印算法

摘要

In order to effectively resist geometric attacks and achieve the copyright protection of digital products,we present a geometric correction-based digital image zero-watermarking algorithm.According to the character of image's invariant centroid that its relative position is stable before and after the geometric attacks,the algorithm selects the invariant centroid of image as the stable reference point,by combining the SIFT point farthest to this reference point,it estimates the geometric transformation parameters by the changes of the position of these two points.In order to eliminate the effect brought by slight errors,after the geometric correction,the blocks'singular value decomposition is car-ried out.The maximal singular values are extracted from each block to form a sequence,then they are transformed into a matrix after the bina-ry quantisation is applied,and the matrix is combined with the watermark to generate the zero-watermarking.Simulation experimental results show that the algorithm can effectively resist geometric attacks with high correction accuracy and strong robustness,and it can also avoid the contradiction between the invisibility and the robustness of the watermark.%为有效抵抗几何攻击，实现对数字产品的版权保护，提出一种基于几何校正的数字图像零水印算法。根据图像不变质心在几何攻击前后相对位置不变的特性，选取图像的不变质心点作为稳定的参考点，结合与其相距最远的SIFT（Scale Invariant Feature Transformation）特征点，利用该两点的变化估计几何变换参数。为消除微小误差带来的影响，在进行几何校正之后进行分块奇异值分解。取各块最大奇异值组成序列，二值量化后变换为矩阵并与水印结合生成零水印。仿真实验结果表明，该算法能有效抵抗几何攻击，校正精度好，鲁棒性强，且能够避免水印鲁棒性与不可见性之间的矛盾。