Efficient Approach for the Localization of Copy-Move Forgeries Using PointRend with RegNetX

Abstract

Digital images are one of the modern era's dominant sources of information and communication. However, an image can simply be altered with the existence of several tools for image editing. These altered images can transmit across platforms of social media to influence some people in society and may have both positive and negative effects. Therefore, the development of technology becomes a necessary issue to detect and localize a forgery in an image. Copy-move forgery (CMF) is one of the most popular forgeries. In CMF, the new forgery image is created by copying a part presented in an image and placing it at a different location on the same image. This paper proposes a PointRend as a technique to localize copy-move forgeries. This work also presents the PointRend framework with a lightweight backbone model RegNetX (PointRend-RegNetX) to detect such forgeries. From the comparative analysis of the proposed technique with ResNet-50 backbone on two standard datasets, it has been shown that the proposed model (PointRend-RegNetX) is superior in MICC F-220 and MICC F-2000 datasets for images that contain copy-move forgeries. In instances segmentation of forged regions, the improved model (PointRend-RegNetX) has achieved a mean average precision (mAP) of 88.5% on MICC F-220 dataset and 86.4 % on MICC F-2000 dataset.