Abstract
dc:descriptionThe main point of this document is to find and test methods that could be used to train neural networks to recognize between AI generated images and real photographs. A new dataset was created for neural network models to train on, which consisted only of landscape images. During neural network model training, three different neural network architectures and four data processing methods were tested to see which works best for the task. Two methods were identified to work the best for this task, achieving an accuracy of 0.88 and 0.85 when classifying unseen data. The results show that the models reached an accuracy that is less than that achieved in other works that use neural networks for a similar task. Furthermore, it was found out that these methods are not universal and do not work for identifying images generated by image generators that the neural network model is not trained on.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zabotka, Mindaugas,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192846566&prefLang=en_US