大尺寸图片的复制粘贴检测

Copy-move detection of large image

针对目前图像复制粘贴检测算法有时间复杂度高、当图像中有光滑区域时错检率大的缺点,将图像角点区域及其八方向邻接圆形区域作为伪造检测区域,提取每个检测区域的矩特征后用k-d树进行排序,并通过比较每个区域矩特征的相似性来定位伪造复制区域。该算法效率高,且对经过旋转、添加噪声、模糊、JPEG压缩等后处理的复制粘贴伪造图像检测效果很好。

Many proposed copy-move forgery detection methods are time consuming, and most of them will be ineffective when there are many smooth regions in a forgery image. This paper used the circular regions around image corners as detection regions, and sorted moments of each detected region by using of k-d tree. It could locate the duplicate regions after comparing the similarities of all detection regions＇ moments. Several experiments demonstrate the efficiency of this method, and also evaluate the copy-move forgery image detecting method＇ s robustness and sensitivity when a forgery image undergoes a serial of image processing such as rotating, adding noise, blurring, compressing etc.