Image tampering detection based on two-branch gating-augmented transformer encoder with attention-based feature fusion

The proliferation of powerful but easy-to-use image editing techniques have led to the spread of image tampering. As a result, image tampering detection has become a key area of research in the field of image forensics. In this paper, we propose an image tampering detection method based on two-branch gating-augmented Transformer encoder with attention-based feature fusion. Specifically, multi-scale RGB and noise features are first extracted by two parallel branches, each branch contains four gating-augmented Transformer encoders and a spatial and channel squeeze & excitation (scSE) module which can enhance the feature representation ability. Then, the extracted features are fused and enhanced via the proposed multi-strip coordinate attention (MSCA) module. Finally, the network makes both imagelevel and pixel-level decisions. Extensive quantitative and qualitative experimental results verify the superiority of the proposed network while ablation study verifies the effectiveness of scSE and MSCA in the network.