Combined SIFT and bi-coherence features to detect image forgery

In this paper, we propose a new method to detect image copy-move tampering using SIFT and bi-coherence features. SIFT feature (scale invariant features transform), which is insensitive to geometrical and illumination distortions, has achieved good results for image region duplication detection as shown in [1], [2] and [3]. However, image deeper forensics analysis such as contents and motives is also important especially when images take the role of legal events. In this paper, we extract matched SIFT keypoints features in a digital image and cluster them based on three-Point Center Clustering (3PCC) method. To estimate the robust geometrical transform parameters between the matched points, Random Sample Consensus (RANSAC) algorithm is adopted. At last, we introduce bi-coherence feature, a normalization higher-order statistic, to register the original source block of the image and the forged one for deeper forensics analysis. Several examples showing the effectiveness of this approach are given.