In the 29th edition of the UEP Foundation’s competition for the best bachelor’s and engineering thesis defended at the Poznań University of Economics and Business in the 2023/2024 academic year, the thesis titled “Detecting Human Faces Generated by Artificial Intelligence Models” won first place. The author of the thesis: Patryk Danielewicz, the supervisor: Prof. Krzysztof Węcel.
The thesis addresses one of the key challenges related to the development of artificial intelligence: recognizing synthetically generated faces. In an era of increasingly advanced generative models capable of creating realistic images of people, the ability to distinguish genuine photos from generated images is of particular importance. This issue has wide-ranging implications, especially in the context of disinformation, digital security, and privacy protection.
As part of the study, a review of the literature on computer vision and the use of neural networks in image analysis was conducted. A crucial stage of the research involved creating a custom dataset containing both real facial photos and images generated by artificial intelligence. On this basis, a comparison of generative and image-classifying models was carried out, fine-tuning neural networks to achieve the best possible effectiveness in detecting fake faces.
The experiment results show that developing an effective classification model is feasible, and the obtained outcomes can serve as a foundation for further research and the development of more advanced methods for detecting synthetic images. The thesis aligns with current challenges in machine learning and generative artificial intelligence, emphasizing the need to develop tools capable of identifying false content in an increasingly digital world.
Congratulations!
It is worth noting that in addition to continuing his studies at the Poznań University of Economics and Business, Patryk Danielewicz is also involved in the OpenFact project and is a member of the board of the SRG Data Science. More information about the competition and photos can be found on the website: fundacjauep.pl