A survey on deep-fake detection algorithms

Since AI technology has been on the rise, applications based in this field are also increasing rapidly. However, some of them are utilizing AI to generate images and videos that display explicit activities with manipulated faces of celebrities or other innocent people, incorporated into them. These images and videos are called Deep Fakes. It causes harm by spreading false information or fake news using social media and other similar applications. Deep fakes are generated using Generative Adversarial Networks also known as GANs and other algorithms which utilize machine learning. However, GANs also perform video deep-fake detection along with Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs). We also used feature extraction to derive basic facial expressions. The best results are obtained using methods based on EfficientNet B7. The accuracy for the state-of-the-art approach in detection is around 88%. Using such mentioned deep learning models, we aim to improve them and increase the accuracy to 93%, with minimal fluctuations to enhance the reliability and robustness of deep fake detection systems.