A Multi-Domain Feature Fusion Framework for Generalizable Deepfake Detection Across Different Generators
Researchers have developed SGFF-Net, a novel framework for detecting deepfakes generated by various models, including diffusion models which pose a challenge for existing methods. This network integrates spatial, gradient, and frequency representations to enhance detection accuracy and robustness across different generation paradigms. Experiments show SGFF-Net achieves high accuracy in intra-dataset evaluations and significantly improves performance in cross-model and cross-paradigm scenarios, especially when combined with data augmentation techniques. AI
IMPACT This framework offers improved generalization for deepfake detection systems, crucial for combating disinformation.