Enhancing Digital Image Forensics: a Comparative Analysis on Deep Learning Frameworks for Copy-Move Forgery Detection

The rapid development of digital image manipulation technologies has made image forgery detection a critical area of research in cyber forensics. Copy-move forgery is one of the most common forms of image manipulation. This involves copying and pasting a region within the same image to hide or duplicate the original content. Traditional detection methods are often ineffective in detecting geometric transformations done on images such as scaling, rotation, and compression. Deep learning technologies have significantly advanced image forensics. However, there is limited research that compares different neural network architectures used for copy-move forgery detection. This study presents a comprehensive analysis of five deep learning models (CNN, GAN, RNN, ResNet50, and VGG16) evaluated on three benchmark datasets to identify their effectiveness in detecting copy-move forgeries. Each deep learning model consists of unique features and capabilities, thus enabling a detailed investigation of their strengths, limitations, and suitability for various manipulation scenarios. Preprocessing steps included normalization, rotation, scaling, and contrast enhancement to ensure robustness. We evaluated the impact of different model parameters on each selected deep learning model. Experimental results show that CNN achieved the highest training accuracy (93.8 %). The VGG16 and ResNet50 outperformed (88 % and 80 % respectively), but not higher than the CNN. The outcomes provide valuable insights about the influence of architectural design on copy-move forgery detection performance. Also contribute toward guiding future forensic model development.