Effective digital image copy-move location algorithm robust to geometric transformations

To make the tampered image more deceivable, rotation and scaling of the duplicated image region, as simple and common image forgery techniques, are often employed. In this paper, a novel approach of copy-move location is presented for copy-move forgery which combines with the keypoint-based method and the block-based method. For a tampered image, we use the split-half recursion matching strategy to match SIFT keypoints. In this step, we obtain the preliminary detection results which indicate the suspected copy-move regions with matching relation. In order to obtain an accurate localization, the proposed method introduces block-based method. First, the affine transformation between two matched regions is estimated. Then, one of the two matched regions is transformed. And then, the ZNCC coefficients are calculated to measure the correlation between the transformed and the matched regions. Experimental results indicate that the presented algorithm can not only localize the tampered regions accurately, but also deal with operations such as rotation, scaling and multiple copy operations. At the same time it also can resist post-processing operations effectively like white Gaussian noise and JPEG compression.