一种基于SIFT特征的抗几何攻击水印算法研究

摘要

The scale invariant feature transform descriptor has been applied to the digital imagewatermarking,but it could lead to mismatch problem when the image has similar regions,so anRIT-SIFT digital image watermarking algorithm is proposed.Before embedding the watermark,we first divide image into blocks,then embed watermark in the low frequency sub-bands respec-tively that come from non-subsample contourlet transform,and it simplifies the watermark em-bedding and increases the embedding capacity.Then we look for local feature points in the originalimage using the SIFT algorithm,fuse the rotation invariant texture feature to SIFT feature vec-tor,and then a more comprehensive description of the image information emerges.It improves thematching accuracy.The experimental results show that the proposed algorithm is effective and hasgood robustness against global geometric attacks,local geometric attacks and the combination ofthe two,as well as in conventional image processing.%尺度不变特征变换描述子已经应用于数字图像水印的构造，但图像存在相似区域时易造成误匹配问题，基于此提出一种改进的RIT-SIFT数字图像水印算法。水印嵌入时先对图像进行分块，然后将水印分别嵌入子块的非下采样Contourlet变换低频子带中，既简化了水印嵌入又增加了嵌入容量。图像匹配时利用SIFT算法在原始图像中寻找局部特征点，将旋转不变纹理特征信息融合到传统SIFT特征向量中，更加全面地描述图像信息，提高了匹配精度。实验结果表明，提出的算法是有效的，对全局和局部几何攻击、组合攻击以及常规攻击均具有很好的鲁棒性。