Digital Forensics: Deep Learning’s Approach to Deepfake Image Identification

Abstract

The paper propose a deepfake detection method leveraging EfficientNet, InceptionResNet, and MobileNet architectures. EfficientNet offers computational efficiency, while InceptionResNet provides depth and feature richness, and MobileNet balances accuracy and computational resources. The models are trained on diverse datasets using transfer learning and fine-tuning for deepfake detection. Enhanced discriminative power is achieved through attention mechanisms and feature fusion, alongside exploration of ensemble methods for improved accuracy. Evaluation on benchmark datasets demonstrates superior effectiveness compared to existing methods. The approach addresses evolving challenges in deepfake detection, showcasing versatility and adaptability for real-time applications.