基于EM算法的人像合成图像的检测  

Detection of Human Composite Images Based on EM Algorithm

作　　者：陈波[1] 朱秋煜[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200072

出　　处：《计算机仿真》2008年第9期201-203,303,共4页Computer Simulation

基　　金：上海市教委重点学科建设项目(J50104)

摘　　要：数字篡改图像即使做的很逼真,没有留下肉眼可见的篡改证据,但图像原先的统计学规律已经改变。以人像合成图像为例,为了使篡改图像更逼真,篡改者需要改变图像大小、旋转图像或拉伸图像的某部分,这些操作需要把原图像重采样到新图像上去。虽然这种重采样通常难以察觉,但图像内部的相关性还是可以作为图像被篡改的证据。这种周期性可以通过EM算法(期望最大化算法)检测出来,从而通过检测合成图像的重采样可作为图像篡改的证据。A digitally altered image,often leaving no visual clues of having been tampered with,can be indistinguishable from an authentic image.The tampering,however,may disturb some underlying statistical properties of the image.Considering the scenario in which a digital forgery is created by splicing together two or more human images.In order to create a convincing match,it is often necessary to re-size,rotate,or stretch the images,or portions of them.This procedure typically requires re-sampling an image onto a new sampling lattice using an interpolation technique.Although a re-sampled image is often imperceptible,it contains specific correlations that,when detected,may represent evidence of tampering.The EM algorithm is recently used for the re-sampling detection.The thesis studies the detection of the re-sampling in image tampering.

关 键 词：期望最大化算法 数字取证 重采样 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]