基于几何均值分解的图像区域复制篡改检测方法  

Detection of Image Region-duplication Forgery Based on Geometric Mean Decomposition

作　　者：赵洁[1,2] 郭继昌[1] 武斌[2] 

机构地区：[1]天津大学电子信息工程学院,天津300072 [2]天津城市建设学院计算机与信息工程学院,天津300384

出　　处：《小型微型计算机系统》2012年第9期2105-2108,共4页Journal of Chinese Computer Systems

基　　金：天津市科技支撑计划基金项目(10ZCKFGX00700)资助

摘　　要：针对现有大多数图像区域复制篡改检测算法提取图像块的特征向量维数较高的缺点,提出一种新的基于几何均值分解的检测算法.将可疑图像分成大小相等的可重叠的子块;并对每个图像块进行几何均值分解并用其表征该子块的特征,形成1维的特征向量;最后对所有的特征向量进行字典排序,并结合图像块的相等位移矢量的发生频率信息,检测并定位出篡改区域.实验结果表明,该算法不仅能够有效检测并定位多区域复制篡改区域,而且对高斯模糊、对比度调整、曝光度调整的后处理操作具有较强的鲁棒性,并且有效地降低了特征向量的维数,提高了检测效率.Aiming at the shortcoming that most of the existing detection algorithms for image region duplication forgery have higher di- mension of image block eigenvectors, a new detection algorithm based on the geometric mean decomposition is proposed. First, the presented method divided the suspicious image into multiple overlapping blocks with the same sizes. Then the geometric mean decom- position is applied to every image block to denote the character and form a one-dimensional eigenvector. Finally, all the eigenvectors are lexicographically sorted and the tampering part is located by means of the displacement vector frequency of every image block. The experimental results show that the algorithm can not only detect and locate multi-region duplication forgery, but also has good ro- bustness to Gaussian blurting, contrast adjustment and exposure adjustment and reduces the amount of eigenvector dimensions effec- tively with improvement of detection efficiency.

关 键 词：区域复制 几何均值分解 图像篡改检测 图像盲取证 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]