Localization and Detection of Copy-Move Forgeries in Post-processed Images Using U-Net

Identification and localization of forgeries in images is of great importance in the field of image processing. Many methods have been introduced to solve this problem. Most provide acceptable results on high resolution images but perform poorly when the resolution is low. Images can be post-processed using techniques such as JPEG compression, brightness change, and the addition of noise to improve the visual quality or create copies for various purposes. Therefore, a new deep learning method is introduced to detect forgeries in images even after post-processing. The proposed model is based on an encoder–decoder architecture designed to learn discriminative features across the boundaries of forged regions. A computer-generated dataset is used for training and evaluation is done using the well-known CoMoFoD dataset. The performance is evaluated using six post-processing techniques, namely brightness change, contrast adjustment, color reduction, image blurring, JPEG compression, and noise addition. To train a robust model for these techniques, data augmentation is employed. The results obtained show that the proposed model outperforms seven recent methods in the literature.