基于随机受限参考区域的合成图像篡改检测  

Detection of Composite Image Based on Random Restricted Reference Area

作　　者：孙鹏[1,2] 王方明 郎宇博 沈喆 单大国 刘磊[1] SUN Peng;WANG Fang-ming;LANG Yu-bo;SHEN Zhe;DAN Da-guo;LIU Lei(Criminal Investigation Police University of China,Shenyang 110035,China;National Engineering Laboratory of Evidence Traceability Technology,Beijing 100038,China;Liaoning Shihua University,Fushun 113001,China)

机构地区：[1]中国刑事警察学院,辽宁沈阳110854 [2]现场物证溯源国家工程实验室,北京110000 [3]辽宁石油化工大学,辽宁抚顺113001

出　　处：《控制工程》2018年第8期1370-1374,共5页Control Engineering of China

基　　金：国家自然科学基金(61307016);现场物证溯源技术国家工程实验室开放课题(2017NELKFKT09);江西省经济犯罪侦查与防控技术协同创新中心开放基金资助课题(JXJZXTCX-014)

摘　　要：针对合成图像篡改取证长期依赖人工经验,容易受到主观因素影响而出现漏判甚至误判的问题,提出了一种基于随机受限参考区域的合成图像篡改区域自动检测方法。首先将图像等分为n？n的子图像块,估计每一个图像子块的光源颜色特征,然后根据篡改区域单连通性的假设构建参考区域,并计算图像子块与参考区域之间的偏色距离,最后,通过自适应阈值对图像中的篡改区域进行分类检测。实验表明,该方法能够自动的检测图像中的伪造篡改区域,为合成图像篡改区域的检测提供了一类科学量化的方法。In order to avoid producing a false positive or false negative judgment in image authentication due to subjective reasons caused by human interaction, we put forward an automated manipulation detection and localization method for composite images based on random restricted reference area. Firstly, the forensic image is divided into n ？n blocks, and the light color features are estimated for each block. Secondly, according to the hypothesis of simple-connected manipulation areas, reference areas are composed of randomly selected blocks and the related blocks. Finally, manipulation areas can be detected automatically in a forensic image by an adaptive threshold after calculating the color shift between each block and reference area. Experiments show that the proposed method can provide quantitative detection of possible splicing manipulation automatically.

关 键 词：图像盲取证 合成图像 光源颜色 自动分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]