Copy-Move Forgery Detection using a Convolutional Neural Network with Channel Attention Module

Copy Move Forgery Detection (CMFD) is an approach utilized to analyze duplicated regions within the same image, which suggests manipulation. It consists of copying a part of the image or similar patches and pasting them within the same image to hide or duplicate objects. A single type of image manipulation is identified as copy-move when significant items or objects are concealed through duplication and reinsertion into in real image. It is difficult to determine whether an image is original. To overcome these challenges, the proposed Channel Attention Module – Convolutional Neural Network (CAMCNN) is employed to recognize image forgeries based on convolutional feature extraction algorithms, such as keypoint and block-based algorithms. This attention mechanism helps to differentiate duplicated areas from similar but non-forged textures. Image resizing and min-max normalization are utilized in the preprocessing stage, while Scale-Invariant Feature Transform (SIFT) is used for feature extraction. The proposed CAM-CNN achieves an accuracy of 97.38% in MICC-F220 dataset, outperforming existing methods such as the lightweight MultiResUNet architecture with Similarity-based Positional Attention Module (SPAM).