A Dual-Network Architecture with Uncertainty-Aware Multimodal Fusion for Deepfake Detection

The detection of AI-generated content, particularly deepfakes, has become a critical issue due to the growing sophistication of generative models. This paper presents MUFDNet, a novel multimodal deepfake detection model that uses paired image and audio data for improved detection accuracy. The model integrates a Vision Transformer (ViT) for image feature extraction and a Graph Neural Network (GGCN) for audio spectrum analysis. By incorporating uncertainty-aware fusion and cross-modal attention mechanisms, MUF-DNet enhances the robustness and generalization of the detection process. Contrastive learning is employed to improve the semantic consistency between the visual and auditory features. Our experiments show that MUF-DNet outperforms existing models in terms of accuracy, precision, recall, and F1 score, demonstrating its potential for deepfake detection in real-world applications.