UnionFormer: Unified-Learning Transformer with Multi-View Representation for Image Manipulation Detection and Localization