Answers are grounded in this paper and cite their sources.
The rapid advancement of deep learning and generative artificial intelligence has enabled the creation of highly realistic artificial images and videos, commonly known as deepfakes. These manipulated media pose significant challenges to digital security, identity verification, privacy, and information authenticity. Consequently, developing reliable techniques for detecting deepfake content has become an important research challenge. This study proposes a deep learning-based approach for detecting manipulated facial images using transfer learning with the MobileNetV2 architecture. The experimental dataset consists of 1,081 real facial images and 960 fake facial images, with the latter obtained through frame extraction from videos available in the FaceForensics++ dataset. The pre-trained MobileNetV2 model is adapted for binary classification by incorporating global average pooling, a fully connected dense layer, dropout regularization, and a sigmoid output layer. The model is trained to classify facial images into two categories: real and fake. Its performance is evaluated using standard classification metrics, including accuracy, precision, recall, F1-score, confusion matrix, Receiver Operating Characteristic (ROC) curve, and Precision-Recall curve. The experimental findings demonstrate the potential of transfer learning for identifying manipulated facial images using a lightweight convolutional neural network architecture. The study highlights the applicability of MobileNetV2 for deepfake image classification under limited-data conditions and identifies dataset diversity and generalization across different sources as important considerations for further research.
No takes yet. Share an insight, caveat, or question.
A 2026 study studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: