A Novel RINBP-DCT-Based Copy Move Forgery Detection Method Against Post-Processing Attacks

Abstract

Digital images are widely used for information sharing, making them increasingly vulnerable to malicious manipulations. Among different types of image tampering, copy move forgery is one of the most common techniques, where a region of an image is duplicated within the same image to conceal or replicate objects. To address this problem, a robust copy move forgery detection method based on Rotation Invariant Neighbors Based Binary Pattern (RINBP) and Discrete Cosine Transform (DCT) is proposed. The proposed technique divides the input image into fixed size overlapping blocks and then calculates the RINBP value for each block and converts it to a DCT vector. Similar blocks are matched by lexicographically sorting all features. Experimental results based on the CoMoFoD and GRIP datasets show that the proposed technique can accurately detect copy move forgeries in a forged image. Furthermore, the method exhibits promising and consistent results both in no post processing (plain copy move) images and in the presence of various postprocessing applications such as JPEG compression, image blurring, color reduction, brightness adjustment, and contrast adjustment. On the CoMoFoD dataset, the proposed method achieved a precision of 0.916 and a recall of 0.946 for plain copy move. On the GRIP dataset, the precision and recall values were obtained 0.969 and 0.859, respectively. Against post processing attacks conducted on both datasets, the method achieved successful results compared to other algorithms.