IFDNet: A Contextually Modulated Deep Network for Robust Image Forgery Detection

The widespread availability of advanced digital editing has made image forgery increasingly sophisticated and harder to detect using conventional techniques. To address this problem, an Image Forgery Detection Network (IFDNet) is proposed. This paper introduced the Dynamic Contextual Modulation Block (DCMB) for enhanced feature representation. Unlike traditional squeeze-and-excitation or Convolutional Block Attention Module (CBAM), the DCMB jointly captures global context and local spatial interactions through adaptive channel-spatial modulation in a single unified block. This enables the network to highlight subtle manipulation artifacts that standard blocks may overlook. The performance of the proposed IFDNet and benchmark models is evaluated on widely used benchmark datasets CASIA V2.0 and MICCF-2000, achieving an average accuracy of 96.53% ± 0.17% and 97.66% ± 0.22% respectively, with an AUC of up to 98.93%, outperforming baseline CNN models. The results demonstrate that our proposed IFDNet model improves robustness and generalization for detecting diverse and complex image forgeries, providing a practical solution for modern digital image forensics.