Image Forgery Detection Using ResNet50, U-Net, and Explainable AI

In this project, I present an explainable deep learning framework for image forgery detection and localization. The system integrates three key components: image-level classification using a ResNet50 model, pixel-level forgery localization using a U-Net segmentation network, and visual explainability through Grad-CAM heatmaps. The goal is not only to detect forged images but also to localize manipulated regions and provide transparent explanations for model decisions. The framework is evaluated on the CASIA 2.0 image tampering dataset and demonstrates reliable classification performance with high forgery recall, meaningful localization, and interpretable visual explanations.