Deepfake Face Extraction and Detection Using MTCNN-Vision Transformers

Deepfake detection is a major problem nowadays. The deepfake detection can be done using face extraction and detection. Strong solutions for face extraction and detection are required in light of the growing prevalence of deepfake technology. This paper combines Vision Transformers with Multi-Task Cascaded Convolutional Networks (MTCNN) to propose a novel method. This approach leverages the real-time face identification skills of MTCNN and the long-range dependency capture ability of Vision Transformers to improve the accuracy of detecting manipulated faces in deepfake footage. We carry out extensive experiments on several deepfake datasets, demonstrating the efficacy of the suggested hybrid strategy. The proposed model findings show that this approach performs better than conventional face identification techniques, particularly when dealing with situations where minor facial alterations are involved. This integration strikes a good compromise between computing efficiency and precision, which makes it a viable option for practical uses. The proposed MT-VIT (Multi-Task Vision Transformer) model provides good accuracy as compared to other state-of-the-art like Residual Networks, Mobile Net, CNN, and Meso-Net.