Abstract: The proliferation of deepfake videos presents a significant challenge to the integrity of digital content. To combat this threat, we propose a novel method for detecting digital deception in videos, termed "Visual Scan." Our approach integrates graph neural networks with convolutional and recurrent neural networks to effectively capture complex relationships within video frames. By leveraging a diverse dataset encompassing various deepfake techniques such as face swapping, voice synthesis, and scene manipulation, our system achieves enhanced robustness and adaptability. Moreover, we introduce a novel adversarial training mechanism to simulate real-world scenarios, enabling our model to effectively counter evolving manipulation strategies. Additionally, our system offers real-time detection capabilities, facilitating the swift identification and containment of manipulated content across online platforms. We anticipate that our approach will significantly improve accuracy levels compared to existing benchmarks in discerning between real videos and deepfakes
Keywords: Digital Deception Detection, DeepFake (DF), Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), Generative Adversarial Networks (GANs), Support Vector Machine (SVM), Long Short-Term Memory (LSTM), ResNeXt.
|
DOI:
10.17148/IJARCCE.2024.134193
