Hybrid visual–forensic deepfake detectionMSc dissertation
A two-stream deepfake image detector that fuses an EfficientNet-B4 network with 406 hand-crafted forensic features through a learned attention gate, evaluated for robustness to JPEG re-compression and on an unseen dataset. On the identity-disjoint FaceForensics++ test set the hybrid reached 0.981 AUC, ahead of either stream on its own.






