Region duplication detection in digital images based on Centroid Linkage Clustering of key-points and graph similarity matching

Region duplication or copy-move forgery is an attack in which a region of an image is copied and pasted onto another location of the same image. In the recent state-of-the-art, a number of key-point based methods have been proposed for copy-move forgery detection in digital images. Though the problems of re-scaling and rotation in region duplication, have been sufficiently investigated using key-point based methods, post-processing based attacks such as flip, blur, brightness and noise, remain an open challenge in this field. In this paper, we address the problem of copy-move forgery detection in images, plus aim to identify copied regions, having undergone different geometric (such as rotation, re-scale), and post-processing attacks (such as Gaussian noise, blurring and brightness adjustment). In the proposed algorithm we introduce a region based key-point selection concept, which is considerably more discriminative than single SIFT key-point extraction. In this work, we apply Centroid Linkage Clustering, to identify duplicated regions in an image, from matched key-points. Also, we introduce a Graph Similarity Matching algorithm, to optimize false matches. Our experimental results demonstrate the efficiency of the proposed method in terms of forgery detection and localization efficiency, for a wide range of geometric and post-processing based attacks in region duplication.