基于SIFT 与 K-means 的图像复制粘贴篡改检测-《计算机技术与发展》

文章信息/Info

Author(s):: YE Yu-qing; QIU Xiao-hui; School of Telecommunications &Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China

摘要:: 针对基于 SIFT 的图像篡改检测算法存在的时间复杂度高、鲁棒性低等问题,提出一种改进的 SIFT 算法。该算法利用 K-means 算法对 SIFT 特征向量矩阵的列向量进行聚类,通过选取与整体差异最小的列向量作为初始聚类中心,最终提取该初始聚类中心所在类的特征向量进行匹配,从而大大降低了特征向量的维数。相比于 PCA-SIFT 算法只能够线性降维,该算法还考虑了局部特征的非线性关系。实验结果表明,与 PCA-SIFT 算法相比,该算法具有更高的检测效率及检测正确率,并对旋转、缩放操作具有较强的鲁棒性。

Abstract:: We propose an improved SIFT algorithm to solve the problem of high time complexity and low robustness for image forgery detection algorithm based on SIFT. It uses K-means clustering algorithm for clustering of column vector of SIFT feature vector matrix. By selecting the column vector with the smallest difference as the initial clustering center,the feature vector of the initial clustering center is extracted,which greatly reduces the dimension of the feature vector. Compared with the PCA-SIFT algorithm can only be linear dimensionality reduction,the algorithm also considers the nonlinear relationship of local features. Experiment shows that compared with PCASIFT algorithm,the proposed algorithm has higher detection efficiency and detection accuracy with robust rotation and scaling operations.