A Multi-Domain Feature Fusion Framework for Generalizable Deepfake Detection Across Different Generators

The article presents SGFF-Net (Spatial-Gradient-Frequency Fusion Network), a multi-domain deepfake detection framework that integrates spatial, gradient, and Discrete Wavelet Transform (DWT)-based frequency representations within a dual residual learning architecture. SGFF-Net achieves 98.95% accuracy in intra-dataset evaluations and significantly improves cross-model and cross-paradigm detection rates, with accuracy enhancements from 70.46% to 79.80% and 69% to 78%, respectively, through multi-source training and data augmentation. This framework's ability to leverage complementary forensic cues across multiple domains enhances robustness against various deepfake generators, providing valuable insights for practitioners aiming to build more effective detection systems.