基于Wasserstein直方图欧氏度量的图像重复区域盲检测算法  

A Blind Algorithm for Duplication Region Detection Based on Wasserstein Histogram Euclidean Metric

作　　者：杜振龙 叶超 李晓丽 沈海洋 宋国美 DU Zhen-long;YE Chao;LI Xiao-li;SHEN Hai-yang;SONG Guo-mei(School of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学计算机科学与技术学院,江苏南京211816

出　　处：《光学与光电技术》2018年第6期25-32,共8页Optics & Optoelectronic Technology

基　　金：国家自然科学基金(61672279);江苏省"六大人才高峰"项目(2012_WLW_023);水文水资源与水利工程科学国家重点实验室开放基金(2016491411)资助项目

摘　　要：提出了一种利用Wasserstein重心坐标作为图像描述子的图像重复区域盲检测算法。通过度量各图像块Wasserstein重心坐标间的最小欧式距离确定图像重复区域初始搜索块,再对初始搜索块进行扩展,利用改进的PatchMatch算法寻找相关度高的图像块,从而搜索出完整的图像重复区域。实验结果表明,提出的基于Wasserstein直方图欧氏度量的图像重复区域盲检测算法能够检测出复制粘贴的伪造区域,对于带有旋转和缩放等几何攻击的伪造区域检测率维持在90%以上。该算法对复制粘贴、尺度变化和旋转几何攻击具有良好的鲁棒性。By using Wasserstein’s concept of the barycentric and barycentric coordinates,a blind region detection algorithm for image region-duplication forgery using Wasserstein barycentric coordinates as image descriptors is proposed.The initial search block of the image is determined by measuring the minimum Euclidean distance between the Wasserstein barycentric coordinates of each image block.The initial search block is extended,and a patch with high correlation is searched for by using the improved PatchMatch algorithm to search for a complete image repetition area.Experimental results show that the proposed algorithm can detect the repeated region by using blind detection algorithm based on the Wasserstein histogram Euclidean metric.The copying and pasting forged regions are also very robust for forged regions with rotation and zoom geometry attacks.

关 键 词：Wasserstein重心坐标 重复区域 块匹配 复制粘贴 几何变换攻击 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]