基于均值漂移的图像复制粘贴伪造盲检测  

Copy-paste image forgery blind detection based on mean shift

作　　者：焦丽鑫 杜振龙[1] 

机构地区：[1]南京工业大学电子与信息工程学院,南京210009

出　　处：《计算机应用》2014年第3期806-809,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61073098);江苏省六大人才高峰基金资助项目(2012-WLW-023);江苏省自然科学基金资助项目(BK2009081);江苏省高校自然科学基金资助项目(09KJB520006;11KJD520007);南京大学软件新技术国家重点实验室开放基金资助项目(KFKT2008B15);东南大学计算机网络和信息集成教育部重点实验室项目(K93-9-2010-04)

摘　　要：针对传统图像复制粘贴伪造盲检测算法存在的耗时长、计算量大、检测精度不高的问题,提出了一种基于均值漂移(MS)的图像复制粘贴伪造盲检测算法。该方法提取图像的加速稳健特征(SURF)特征点,通过最近邻匹配方法进行特征匹配,滤除冗余点,初步定位复制粘贴伪造区域。利用均值漂移将具有相同或相似属性的图像像素分割为同一区域,借助匹配后的SURF特征点与其所在均值漂移分割区域的位置依赖关系确定伪造区域,并采用边缘直方图和HSV颜色直方图衡量特征点所在分割区域与相邻区域间的相似度,将大于相似度阈值的邻域划分到复制粘贴伪造区域中,进一步细化伪造检测结果,最终实现图像的复制粘贴伪造盲检测。实验结果表明,在细节轮廓清晰和灰度值变化明显的图像中,该算法能够达到比较理想的检测效果,能够鲁棒地、高效地检测出图像的复制粘贴伪造区域。The traditional blind detection methods of image copy-paste forgery are time consuming, of high computation cost and low detection precision. A blind detection algorithm of copy-paste image forgery based on Mean Shift （MS） was proposed in this paper, which extracted Speed Up Robust Feature （SURF） points and then performed feature matching utilizing the method of best bin first in order to filter redundant points and locate the copy-paste forgery regions preliminarily. Pixels with the same or similar attributes would be segmented in the same region after implementing MS. The copy-paste regions could be detected according to the position dependency between matched feature point with its segmented region of MS and the detection result would be further refined by comparing the similarity of edge histogram and HSV （ Hue-Saturation- Value） color histogram among the segmented regions of matched SURF and its neighborhood, and those regions with large similarity were included in the forged region. The experimental results show that the copy-paste forgery regions are detected accurately in the image with clear outline and rich details, and the proposed algorithm can robustly and efficiently detect the copy-paste forgery regions of image.

关 键 词：数字图像 复制粘贴 图像伪造 加速稳健特征 均值漂移 

分 类 号：TP391.413[自动化与计算机技术—计算机应用技术]