A Comprehensive Overview of Deep Learning for Deepfakes: Generation, Detection, Dataset: A Survey | ||||
| IJCI. International Journal of Computers and Information | ||||
| Article 6, Volume 12, Issue 2, July 2025, Page 60-80 PDF (1.97 MB) | ||||
| Document Type: Original Article | ||||
| DOI: 10.21608/ijci.2025.330774.1180 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Eman AbdElfattah  1; Nader Mahmoud2; Hamdy M.Mousa 1; Ashraf Elsisi3
| ||||
| 1Computer Science Department, Faculty of Computers and Information, Menoufia University, Egypt | ||||
| 2Cybersecurity Department, Engineering and Information Technology College, Buraydah Private Colleges. | ||||
| 3Faculty of Computers and Information, Menofia University, Egypt | ||||
| Abstract | ||||
| The rapid evolution of deep learning techniques, particularly through generative adversarial networks (GANs), has enabled the creation of hyper-realistic synthetic media, heightening concerns in domains such as politics, entertainment, and security. Consequently, there has been a heightened interest in developing robust deepfake creation and detection systems. This paper provides a comprehensive survey of state-of-the-art deepfake detection methodologies, more specifically we focus on video-based, and image based approaches and their applications. It seeks to enhance the reader's understanding of recent developments, vulnerabilities in existing security measures, and areas for improvements. This research is among the few studies that extensively review datasets, highlighting their strengths and weaknesses. Which are crucial for training and validation purposes of deepfake models, such as: FaceForensics++, DeepFake Detection Challenge (DFDC), and Celeb-DF, emphasizing the need for diverse and realistic datasets to ensure robust model generalization. It thoroughly addresses every aspect of identifying and creating deepfakes, providing the reader with a comprehensive understanding of these topics in a single study. Recent studies primarily use Convolutional Neural Networks (CNNs) for deepfake detection, with the main goal of optimizing detection accuracy, also focuses on comparing different approaches. From previous studies we conduct that XceptionNet, EfficientNet, and I3D are top models for deepfake detection, each excelling in different areas: XceptionNet identifies subtle manipulations, EfficientNet offers balanced and efficient video frame analysis, and I3D specializes in real-time video sequence processing. | ||||
| Keywords | ||||
| Deepfake; Detection; Generation; CNN; RNN | ||||
|
Statistics Article View: 266 PDF Download: 117 |
||||
