Abstract
Copy-move forgery represents one of the most prevalent forms of digital image manipulation, where regions from the same image are copied and pasted to conceal or duplicate objects. This study presents a comprehensive analysis of feature-enhanced deep learning algorithms for detecting copy-move forgeries in digital images. Our research investigates the effectiveness of various deep learning architectures including Convolutional Neural Networks (CNN), ResNet, and Vision Transformers (ViT) combined with traditional feature extraction methods such as Scale-Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), and Local Binary Patterns (LBP). The experimental evaluation was conducted on multiple benchmark datasets including MICC-F2000, CoMoFoD, and COVERAGE, comprising over 15,000 images with varying levels of post-processing operations. Results demonstrate that the proposed feature-enhanced CNN-ResNet hybrid model achieves superior performance with 96.7% accuracy, 95.2% pr