SIFT match removal and keypoint preservation through dominant orientation shift

In Image Forensics, very often, copy-move attack is countered by resorting at instruments based on matching local features descriptors, usually SIFT. On the other side, to overcome such techniques, smart hackers can try firstly to remove keypoints before performing image patch cloning in order to inhibit the successive matching operation. However, keypoint removal determines per se some suspicious empty areas that could indicate that a manipulation has occurred. In this paper, the goal to nullify SIFT matches while preserving keypoints is pursued. The basic idea is to succeed in altering the features descriptor by means of shifting the dominant orientation associated to a specific keypoint. In fact, to provide rotation invariance, all the values of the descriptor are computed according to such orientation. So doing, it should impair the whole matching phase.