Abstract
The accelerated growth in synthetic visual media generation and manipulation has now reached the point of raising significant concerns and posing tremendous intimidations towards society. There is an imperative need for automatic detection networks towards false digital content and avoid the spread of dangerous artificial information to contend with this threat. The existing DeepFake detectors do not work well on these unseen, newly generated datasets.
In this thesis, we propose a novel DeepFake detection approach based on Supervised Contrastive Learning. In addition, we propose a fusion model by combining weighted scores from Supervised Contrastive Learning and Xception network. We successfully achieve superior performance compared to the State-of-The-Art deep networks. Our fusion model consistently outperforms Supervised Contrastive Learning model and Xception model on the known data. We also provide extensive evaluations on the State-of-The-Art networks using the hand-crafted features and deep features as benchmarks.
In contrast to most of the earlier works, the thesis also focuses on generalization of DeepFake to detect unknown attacks by training on known attacks. The obtained results outperform the current approaches for detecting unknown DeepFakes.
The thesis also aligns with reproducible research by making the code available online.