Research Article

Semantic Modeling and Pixel Discrimination for Image Manipulation Detection

Figure 3

The outputs of our approach. The first row indicates the source image, the manipulated images, and the ground truth. The second row demonstrates the detection result of forgery object-level branch (), the pixel-level branch with the feature fusion module (), and the two branches with the integrated fusion module ().