Vision Transformer with Attention Mechanism for Copy-Move Forgery Detection

This paper presents a novel approach to detecting and localizing copy-move forgery (CMF) in digital images using the BusterNet model enhanced with Vision Transformer (ViT) technology. Recognizing the limitations of traditional convolutional neural networks (CNNs) in understanding the broader context of images, our study integrates the transformative capabilities of ViTs, renowned for their proficiency in capturing global dependencies through self-attention mechanisms. This hybrid model employs a dual-branch architecture focusing on both manipulation detection and similarity analysis to robustly address varied forgery scenarios. Extensive tests demonstrate that the BusterNet with ViT performs better than conventional methods, providing high precision, recall, and intersection over union (IoU) scores. These results underscore the model's efficacy in detecting sophisticated forgeries that conventional methods might miss, highlighting its potential for advancing digital image forensics.