一种基于SD-K-means++聚类的图像复制粘贴篡改检测方法研究  

Research on Image Copy and Paste Tamper Detection Method Based on SD-K-means++ Clustering

作　　者：陈宇川 CHEN Yuchuan(College of Software,Jilin University,Changchun Jilin 130015,China)

机构地区：[1]吉林大学软件学院,吉林长春130015

出　　处：《信息与电脑》2025年第5期136-139,共4页Information & Computer

摘　　要：文章提出了一种基于SD-K-means++聚类的图像复制粘贴篡改检测算法。传统K-means++算法在进行聚类算法之前,仍需指定聚类的数目。强制指定聚类数目可能将数据过度简化或错误分类,且不能对复杂的聚类形状进行有效处理。文章所提算法通过引入每个聚类的平方误差和、轮廓系数以及戴维森堡丁指数,对聚类的数目进行最佳估计,再选出一个特征点作为第一个聚类中心,不断进行迭代,直到最佳估计的聚类数目,最后进行估计仿射变换,排除不相关的点,如检测到两个及以上的聚类数量则判定为受到篡改。实验结果表明,该算法具有较强的适应性,相比传统算法在误匹配率上有显著改进。The article proposes an image copy-paste tampering detection algorithm based on SD-K-means++clustering.Traditional K-means++algorithms still need to specify the number of clusters before performing the clustering algorithm.The mandatory specification of the number of clusters may oversimplify or misclassify the data and does not allow efficient handling of complex cluster shapes.The algorithm proposed in the article provides the best estimation of the number of clusters by introducing the sum of squared errors,profile coefficients,and Davidson’s Fortin index of each cluster,then selects a feature point as the first cluster center,and iterates until the best estimated number of clusters,and finally performs the estimation affine transformation to exclude irrelevant points,and determines that it is subjected to tampering if it detects two or more numbers of clusters.After experimentation,it can be seen that the algorithm is highly adaptable and has significant improvement in the false matching rate compared to the traditional algorithm.

关 键 词：SD-K-means++算法 SIFT算法 复制粘贴篡改 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]