DeepFakeGuard: Real-Time Deepfake Video Detection Leveraging Celeb-DF Dataset and CNN-LSTM Framework
S R Viknesh, Praveen Kumar M, A. G., G. Sivakarthi · 2025
Deepfake technology has significantly eroded the veracity of digital media across the world, raising concerns of misinformation and media manipulation. To counteract this, we have developed DeepFakeGuard, a powerful deepfake detection system with robust deep learning algorithms present on the web. The system works to identify deepfake videos by checking each subsequent frame for irregularities, for instance, unbalanced lip movements, abrupt light changes, or unnatural cuts. DeepFakeGuard has been pre-trained on a wide variety of datasets primarily focused on the Celeb-DF dataset with high-quality and challenging-to-detect deepfake content. This provides the system with flexibility and precision for multiple applications such as short-form broadcast, high-definition video streams, and live streams. The initial step was preprocessing of data meticulously with focus on neatness and elimination of data noise to ensure the model works at its best. We then employed the latest deep learning architectures, a combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), to detect the slightest disparities between subsequent frames. Our models have been exhaustively tested to identify tampered content from real media at high accuracy levels. The web-based system provides real-time detection of deepfake, with the ability to upload media files to be detected in real-time. In addition, its robust backend infrastructure for processing video prevents disruption in smooth processing, making DeepFakeGuard deployable on a wide variety of environments such as content moderation, digital forensics, and media verification. The result of our work is evidence of how DeepFakeGuard can enhance digital media security through effective deepfake content detection.