UMIF-Net: unified framework for multi-type image forgery detection and localization

Multi-type image forgery detection and location is an important challenge in digital forensics. The existing methods are difficult to fully take into account the detection requirements of forgery types such as copy-move, removal, and splicing. To this end, this paper proposes a multi-type image tampering detection and localization network (UMIF-Net) by imitating the process of human identification of forged images. Firstly, according to the forgery type and implementation mechanism, the common features of different forgery types are extracted. The difference features are extracted by the similarity detection branch and the splicing detection branch. The dual attention mechanism is used to fuse the common features and the different features. In the training phase, the joint loss function is used to optimize the network parameters. Experiments on six public datasets show that UMIF-Net is superior to the existing mainstream methods in AUC and F1 scores, especially on the Coverage dataset, the positioning performance is improved by more than 10 %. In addition, the fine-tuning experiments of the model on different data sets further verify its superior generalization and adaptability to multiple forgery types. UMIF-Net provides an efficient solution for the field of image forgery detection, showing significant robustness and reliability in complex and diverse forgery scenarios.