Comparative Analysis of Deep-Fake Detection Methods

The easy availability of deepfakes because of advancements in generative adversarial networks (GANs) and other deep learning models, has led to popularization of extremely realistic fake media which can easily be misused. This review paper gives a comparative analysis of different deepfake detection methods in images as well as videos format, it focusses on advanced methodologies like hybrid deep learning, surface anomaly detection, vision language adoption, and transfer learning. Different methods like CNN-LSTM, GAN based approaches and architectures like CLIP, have different strengths in deepfake detection. Each method is assessed for its effectiveness in different real-world scenarios, generalization capabilities and performance with multiple datasets. We will also look into the shortcomings in every method. This analysis aims to find optimal deployment of these techniques and find which technique is optimal for detecting deepfakes to combat misuse and misinformation caused by such altered images and videos.