Deepfake Detection From Face-swapped Videos Using Transfer Learning Approach
Md. Tahmid Hasan Fuad, Faiyaj Bin Amin, Sk. Md. Masudul Ahsan · 2023
Deepfakes are synthetic media created using artificial intelligence and machine learning techniques. Deepfakes are produced by employing a generative model to alter photos, videos, or sounds and then producing a new piece of media that mimics the original. With improvements in AI technology and the availability of significant computer resources, deepfake production has become increasingly feasible. Although there are many potential uses for deepfakes in the domains of entertainment, art, and research, they also present significant ethical and security issues. Deepfakes have the potential to influence people and spread false information, which could have detrimental effects on both the individual and the larger society. To stop malicious exploitation, it's crucial to create methods for spotting deepfakes and to control their use. This paper focuses on proposing a model for detecting deepfake videos with higher accuracy on a created dataset and the existing state-of-the-art dataset. A transfer learning based model has been proposed for deepfake detection. Wide ResNet and CNN have been implemented in the proposed model. The proposed model has been tested on both the created dataset of 121 videos and 3762 videos from Deepfake Detection Challenge dataset and achieved 83.47% and 82.4% accuracy respectively which is better than other pretrained models. High computational requirement has been one of the major challenges of this work.