| Improving DeepFake Video Detection Performance with a Noval Deep Learning Approach |
| Paper ID : 1090-ICEEM2023 |
| Authors |
|
Mona A. Fouda *1, Walid El-Shafai2, El-Sayed M. EL-Rabaie1 1Department of Electronics and Electrical Communications Engineering Faculty of Electronic Engineering, Menoufia University Menouf 32952, Egypt 2Security Engineering Lab, Computer Science Department Prince Sultan University Riyadh 11586, Saudi Arabia |
| Abstract |
| with the increase, in both the quantity and quality of deepfake videos it has become crucial to have detection systems that can alert users on social media and the internet about potentially misleading content. Despite advancements in algorithms, software and smartphone apps of creating manipulated videos and swapping faces automated systems for detecting face forgery in videos still have limitations. These systems often show bias towards the dataset used to train them. Our research paper addresses this issue by proposing an approach for detecting deepfake media. We introduce a custom Visual Geometry Group (VGG16) deepfake approach that combines VGG16 with convolutional neural network architectures. To evaluate our method we use the deepfake detection challenge on Kaggle (DFDC) dataset to build network models and compare our custom approach with VGG16. Additionally we analyze how data augmentation techniques impact the performance of Convolutional Neural Network (CNN) based deepfake detectors examining this effect on both VGG16 and our custom VGG16 approach utilizing the DFDC dataset. Our novel research approach achieves a high level of accuracy, with a precision of 0.983, a recall of 0.975, an f1-score of 0.979, and 0.986 accuracy for deepfake detection. Overall, this study presents a promising approach for improving the accuracy of DeepFake video detection, which is an important step towards mitigating the potential negative impacts of DeepFake technology. |
| Keywords |
| Deepfake, Deep learning, CNN, Data Augmentation, Deepfake Detection, and VGG16. |
| Status: Accepted |