基于暗通道的模糊型变造区域的定量鉴定算法  

Copy-Paste with Edge Blur Tampering Region Authentication Based on Dark Channel

作　　者：杜振龙 蒋子贤 王姣姣 李晓丽 Du Zhenlong;Jiang Zixian;Wang Jiaojiao;Li Xiaoli(School of Computer Science and Technology,Nanjing Tech University,Nanjing 211816)

机构地区：[1]南京工业大学计算机科学与技术学院,南京211816

出　　处：《计算机辅助设计与图形学学报》2020年第9期1466-1472,共7页Journal of Computer-Aided Design & Computer Graphics

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).

摘　　要：针对复制粘贴常伴随着边缘模糊变造制作伪造图像,提出了基于暗通道的边缘模糊复制粘贴篡改区域鉴定算法.图像的暗通道属性可作为区分清晰图像和模糊图像的自然度量,依据边缘模糊区域的暗通道非零像素数高于其他区域,快速定位至粘贴物体的边缘模糊区域;对图像进行归一化分割,形成语义较完整的候选鉴定区域集;在希尔伯特空间描述区域集,通过区域间的归一化相关系数确定复制源区域.对从因特网搜集的室外图像集进行实验,以源复制区域为算法检测到的区域比较参考,实验表明所提算法平均检测准确率提高了约6%,检测结果不受光照强度、噪声和JPEG压缩率的影响.Copy-paste with edge blurring manipulation is the commonly used tampering trick,and the generated forgery image is difficult to authenticate.A tampering image authentication algorithm based on dark channel is proposed,it focuses on detecting the forgery image generated by copy-paste with edge blurring manipulation.Dark channel is a novel image attribute.Because the number of nonzero pixels in blur region is higher than the one of intact regions,dark channel is exploitable for rapidly distinguishing the tampering region.The normalized image segmentation is applied for generating the semantic regions,these regions are described in Hilbert space,and their normalized correlation coefficients are used for determining the copy-paste source.Experiments demonstrate that the proposed approach can increase the average detection accuracy 6%,regardless of illumination variation,noise and JPEG compression ratio change.

关 键 词：暗通道 复制粘贴 边缘模糊 归一化分割 希尔伯特空间 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]