Deep Fake and Digital Forensics

Deepfakes have posed a significant challenge to digital forensics, and there is an increasing need for high accuracy deepfake (DF) detection models in real-world scenarios. This research examines and fine-tunes the MesoNet model to improve its performance on a large dataset of 140K authentic and manipulated images. The original MesoNet model achieved an accuracy of 87.1%. However, after fine-tuning and optimizing the model’s weights, the accuracy improved to 96.20%. This was accompanied by a sensitivity of 97.48% and a specificity of 94.75%, indicating that the model is highly effective at detecting genuine images and accurately identifying forged ones. This research contributes to the advancement of DF detection mechanisms in real-world scenarios.