A Survey on AI-based Deep Fake Detection for Human Face Images and Videos

Abstract

Technology for making and changing multimedia stuff has come a long way, and now we can create visuals that look really real. DeepFake tech uses these deep learning models to mess with faces or make new ones, and it’s so good that telling real from fake is tough sometimes. I think that’s part of why it’s exciting but also scary. There are good sides to it, like in movies or TV shows where they improve effects, or even video games to make things look better. But people are using it badly too, for spreading else info or pretending to be famous people, which can cause big problems. To fight that, researchers are working on ways to detect DeepFakes, mostly with deep neural networks. Basically, DeepFakes are just media that’s been changed or created by training these models to swap or add visual bits, especially faces. This paper looks at different detection methods for images and videos of faces, sorting them by how they detect, what techniques they use, and how well they work in tests. It also covers how DeepFakes are made in the first place, putting them into five main groups. I am not totally sure about all the details there, but it seems important to understand both sides. Datasets for DeepFakes are another thing they review, looking at what’s common and how theyve gotten better or more varied lately. One big issue is making detection models that work on new kinds of fakes they haven’t seen before, which sounds tricky. Overall, the survey points out challenges in creating and spotting these things, and some open questions that need more work. Hopefully, this helps push forward better deep learning ways to catch DeepFakes in faces and videos, though it might take time to get really reliable.