Deepfake Detection

Title Deepfake Detection
Summary Detecting deepfake images and videos using a diversified ensemble of deep models
Keywords deepfake, deep learning, generative models, ensemble, diversity
TimeFrame Fall 2021
References 1- Tolosana, Ruben, et al. "Deepfakes and beyond: A survey of face manipulation and fake detection." Information Fusion 64 (2020): 131-148.

2- Liu, Xin, and Xiao Chen. "A Survey of GAN-Generated Fake Faces Detection Method Based on Deep Learning." Journal of Information Hiding and Privacy Protection 2.2 (2020): 87.

3- Hsu, Chih-Chung, Yi-Xiu Zhuang, and Chia-Yen Lee. "Deep fake image detection based on pairwise learning." Applied Sciences 10.1 (2020): 370.

4- Khodabakhsh, Ali, et al. "Fake face detection methods: Can they be generalized?." 2018 international conference of the biometrics special interest group (BIOSIG). IEEE, 2018.

5- Mashhadi, Peyman Sheikholharam, Sławomir Nowaczyk, and Sepideh Pashami. "Parallel orthogonal deep neural network." Neural Networks 140 (2021): 167-183.

Prerequisites Good knowledge of machine learning and deep learning; programming skills for implementing deep learning algorithms
Supervisor Stefan Byttner, Jens Lundström, Peyman Mashhadi
Level Master
Status Open

Fake images and videos, including facial information manipulation and generation, particularly with the rise of deep learning methods, have become a massive concern in recent years. The spread of disinformation causes a wide range of consequences ranging from individual abuses to political manipulations. Therefore, the need for detecting fake information is of crucial importance. The importance is to the extent that even in data science competition platforms like Kaggle, some giant companies, including Amazon, Google, and Facebook, co-sponsored a competition in 2020 for deepfake detection.

One of the most influential, widely used methods for generating fake images and videos using deep learning models is Generative Adversarial Networks (GANs). Different variations of GANs have gained tremendous abilities in generating realistic images and videos. This thesis aims to develop deep learning methods to detect face-generated content using deepfake methods. To this end, one promising direction is to encourage diversity in a deepfake detection system. The hypothesis is that a model with a certain level of diversity would be harder to be tricked against fake content.