Published Paper Details:

The rapid advancement of Artificial Intelligence (AI) and generative deep learning models has significantly enhanced the capability to produce highly realistic synthetic images, commonly referred to as deepfakes. Although these technologies have enabled innovations in entertainment, healthcare, and digital content creation, they have also introduced critical challenges related to misinformation, identity theft, cybercrime, and digital media authenticity. Existing deepfake detection approaches predominantly rely on image-based analysis using Convolutional Neural Networks (CNNs), yet their performance often deteriorates when confronted with unseen datasets, sophisticated generative models, adversarial attacks, and manipulated metadata. To address these limitations, this research proposes a novel multi-modal deepfake image detection framework that integrates visual feature learning through Vision Transformers (ViTs) with fake metadata analysis to improve detection robustness and generalization. The proposed framework combines EfficientNet-based feature extraction, attention mechanisms, Vision Transformers for global contextual representation, and metadata forensic analysis to simultaneously examine image content and associated metadata. This multi-modal strategy enables the detection of subtle spatial inconsistencies, semantic artifacts, and metadata anomalies that may remain undetected when relying solely on visual information. The model is trained and evaluated using incorporating benchmark deepfake datasets, comprehensive preprocessing, data augmentation, transfer learning, and ensemble classification techniques. Performance is assessed using standard evaluation metrics including accuracy, precision, recall, F1-score, Receiver Operating Characteristic (ROC) curve, and Area Under the Curve (AUC), along with cross-dataset validation to evaluate model robustness under diverse real-world conditions. The expected outcome of this research is the development of a scalable, reliable, and computationally efficient deepfake detection framework capable of accurately identifying AI-generated images while maintaining strong generalization across different manipulation techniques. By integrating visual and metadata-based forensic evidence, the proposed system enhances digital image authentication and contributes to combating misinformation, strengthening cybersecurity, supporting digital forensic investigations, and preserving trust in digital media. The research advances the field of computer vision by presenting a comprehensive multi modal detection framework suitable for next-generation AI-generated image verification systems.