A Forensic Scheme for Revealing Post-processed Region Duplication Forgery in Suspected Images

Abstract

Recent researches have demonstrated that local interest points alone can be employed to detect region duplication forgery in image forensics. Authentic images may be abused by copy-move tool in Adobe Photoshop to fully contained duplicated regions such as objects with high primitives such as corners and edges. Corners and edges represent the internal structure of an object in the image which makes them have a discriminating property under geometric transformations such as scale and rotation operation. They can be localised using scale-invariant features transform (SIFT) algorithm. In this paper, we provide an image forgery detection technique by using local interest points. Local interest points can be exposed by extracting adaptive non-maximal suppression (ANMS) keypoints from dividing blocks in the segmented image to detect such corners of objects. We also demonstrate that ANMS keypoints can be effectively utilised to detect blurred and scaled forged regions. The ANMS features of the image are shown to exhibit the internal structure of copy moved region. We provide a new texture descriptor called local phase quantisation (LPQ) that is robust to image blurring and also to eliminate the false positives of duplicated regions. Experimental results show that our scheme has the ability to reveal region duplication forgeries under scaling, rotation and blur manipulation of JPEG images on MICC-F220 and CASIA v2 image datasets.