A Hybrid U-Net Approach to Digital Image Forgery Detection

Abstract

Abstract To detect and localise picture modifications, image forgery detection is vital study. This work introduces a hybrid U-Net-based image forgery detection method that blends deep learning with semantic segmentation models. A hybrid U-Net architecture with feature extraction, semantic segmentation, and classification modules is used in our method. Feature extraction uses the VGG16 network, whereas semantic segmentation uses a modified U-Net design with residual connections. The classification module detects picture modifications using binary classification on a fully linked network. We tested our method on the CASIA2 dataset, which contains 10,000 photos with various image alterations. We tested our strategy using 5-fold cross-validation and compared it to several state-of-the-art methods. Our method outperformed others in accuracy, robustness, and efficiency, showing its promise for identifying image modifications in real-world circumstances. Our effective and efficient method for identifying diverse picture modifications with high accuracy and robustness makes a substantial addition to image forgery detection. Digital forensics, picture authentication, and related industries will benefit from the suggested technique, which will make image-based systems more trustworthy.