Transfer Learning Based Detection of Artificial Intelligence-Generated Human Faces

This study aims to detect human face images generated by artificial intelligence. A balanced dataset containing real and artificial intelligence-based human face images was created. Thirty thousand images were used to train an EfficientNet-B0 model through transfer learning. The performance was evaluated using a Leave-One-Domain-Out analysis. This analysis was performed to evaluate the model’s generalization performance across unseen data domains. Experimental results demonstrate that the model achieved a promising accuracy across different data sources. The proposed model achieved an average accuracy of 0.82 and an F1-score of approximately 0.85 in the K-fold cross-validation setting based on out of fold predictions. This research contributes to the literature by providing a practical deep learning-based approach for detecting artificial intelligence-generated synthetic human faces.