Self-Attention Forged Images Detection Network with Three-Branch Multi-Feature Fusion

Images play a crucial role in modern society with their intuitiveness and authenticity. However, with the rapid development of generative models, forging realistic images has become easier. The forged images not only mislead public perception, but may also pose serious risks to news authenticity or social stability. In this work, we propose an image forgery detection method based on Swin Transformer, which extracts spatial domain (raw image), frequency domain (discrete cosine transform), and time-frequency domain (wavelet transform) features of the image, and introduces multi-head attention mechanism in the feature fusion stage to achieve deep fusion of cross-domain features. The experimental results demonstrate that the proposed method achieves excellent detection performance and offers a novel approach for image forgery detection.