Keypoint-Based Detection and Region Growing-Based Localization of Copy-Move Forgery in Digital Images

Abstract

The issues caused due to image manipulations are common these days. Furthermore, it causes severe troubles in news broadcasting, social media, and digital media forensics. Mainly, the types of image manipulations are divided into four; they are splice forgery, copy-move, morphing, and retouching. Among them, copy-move forgery is one of the most challenging manipulations to detect since it does not change image characteristics while performing the copy-move forgery operation. In this paper, we propose a copy-move forgery detection and localization scheme that detect forgery regions in the image even though the forgery image undergoes translation, scaling, and rotation attacks. The scheme uses the SIFT algorithm for keypoints extraction from the forgery image, DBSCAN to cluster these keypoints, and Hu’s invariant moments are used to identify similarity between two suspicious regions in the image. Lastly, a region growing is performed around these detected regions to localize copy-move forgery regions. The scheme has experimented with a CoMoFoD dataset which is publicly available, and the result shows that the proposed scheme outperforms the state-of-the-art non-deep learning-based copy-move forgery techniques in terms of recall, FNR, F1-score, and also in computational time.