Enhancing Deepfake Detection: Leveraging Deep Models for Video Authentication
Vikram Sundaram, B. MUTHU SENTHIL, Susmitha Vekkot · 2024
This research endeavors to revolutionize deepfake detection by achieving unparalleled accuracy with a modest dataset. Leveraging CNN-based feature extractors and either CNN or MLP as high-level architectures, the methodology prioritizes precision in discerning authentic from fabricated content. Remarkably, both the InceptionV3 and InceptionResNetV2 models, employing Multilayer Perceptrons (MLP) as their highlevel architecture, demonstrate outstanding accuracy rates, of $\mathbf{84.6\%}$. Moreover, the Receiver Operating Characteristic (ROC) for InceptionResNetV2 stands at an exceptional 0.99, while for NasNetLarge, it reaches $\mathbf{0. 9 8}$. Additionally, the MLP InceptionResNetV2 model demonstrated exceptional performance with 12 True Negatives (TN) and only 1 False Negative (FN). These findings underscore the efficacy of the proposed approach in combating the proliferation of deceptive media content.