针对Poisson方程图像编辑的篡改检测

给出一种针对Poisson图像编辑方法的被动认证方法.首先通过Poisson方程对图像分块粗略匹配以快速地检测出两个相似的克隆区域,再利用像素级精确匹配得到对应区域的精确边界.为了进一步推测篡改者的动机,估计原始图像的真实面貌,将两个对应区域边界上的Laplace值进行累加,比较两个相似区域在边界处的不连续性,可以确定边界具有较大Laplace累加值的区域为克隆目标区,从而实现对Poisson图像编辑的被动认证.实验结果表明,该方法能对同一幅图像中的克隆编辑进行可靠检测,并确定源区和克隆区,为数字图像的被动认证提供有效手段.

Image Authentication against Local Area Cloning by Detecting Poisson Editing

An image authentication technique is proposed to detect image forgery that is done by Poisson editing,which copies a small region from the same image and makes the image look natural.By roughly matching the regions which satisfy the Poisson equation,two similar image blocks,one of which is a cloned result of the other,can be identified quickly.The precise boundaries of these two areas are figured out by refined matching with pixel-level precision.Laplacian values on the two boundaries are summed up respectively,and the original and cloned regions can be identified according to the extent of image discontinuity in the boundary regions.The region with a larger cumulated Laplacian value can be judged as the target area.Provided in this paper is a means for passive image authentication against cloning within the same image.