Cyber Security Focused Deepfake Detection System Using Big Data

Cyber Security Focused Deepfake Detection System Using Big Data

연구 분야: Databases

학회: SN Computer Science

In recent months, the surge in free deep learning-based tools has facilitated the seamless creation of convincing face exchanges in videos, leading to the emergence of “DeepFake” (DF) videos. While digital video manipulations have existed for decades, recent advances in deep learning have significantly heightened the realism of fake content, making the creation of AI-synthesized media, commonly referred to as DF, a straightforward task. However, the detection of these DeepFakes remains a formidable challenge, acknowledging the complexity of training algorithms to identify them. In alignment with innovative approach, Big Data and Cybersecurity principles is leveraged to fortify system. Employing Convolutional Neural Network (CNN) at the frame level for feature extraction, Big Data techniques for comprehensive data collection and analysis is utilize. The Recurrent Neural Network (RNN) is then trained to discern temporal inconsistencies introduced by DF creation tools. This model strikes an optimal balance between computational efficiency and detection accuracy, outperforming competitors such as ‘FakeCatcher,’ ‘XceptionNet,’ and ‘Face X-ray.’ Demonstrating resource-efficiency over high-resource alternatives, the hybrid ResNext 50 + LSTM architecture emerges as a superior choice for real-world applications, addressing the formidable challenge of identifying increasingly convincing DF content.