GTA-Net: A Robust Method for Deepfake Face Image Detection

The rapid advancement of artificial intelligence technology has resulted in the emergence of deepfake, which has had a significant impact on various fields due to its realistic effects. Addressing the challenges posed by deepfake has become a crucial area of research. In this study, we propose a two-stream network framework, GTA-Net, for the detection of deepfake face images. The framework comprises a Global Residual Attention module (GRA), a Texture Feature Saliency module (TFS), and an Attention Feature Fusion module (AFF). We incorporate Local Binary Patterns (LBP) features into the network input to guide the decision-making process of the model towards prominent texture features, thereby enhancing detection accuracy. Additionally, we employ a residual attention mechanism to focus on specific deepfake-generated features and improve robustness by avoiding interference from content-preserving manipulations. Experimen-tal results show that the proposed method has high detection accuracy and strong robustness against degraded datasets, and generalization for cross-dataset detection.