University of Twente Student Theses
StyleDemorpher : High-Quality Face Demorphing Using StyleGAN2's Latent Space
Ismayilov, Raul (2024) StyleDemorpher : High-Quality Face Demorphing Using StyleGAN2's Latent Space.
This is the latest version of this item.
|
PDF
9MB |
| Abstract: | Morphing attacks pose a substantial threat to Automated Border Control (ABC) systems by allowing identity documents to be used by multiple individuals, compromising border security. Face demorphing has become crucial for countering these attacks, aiming to disentangle and reconstruct hidden identities within a morph. This paper introduces a novel face demorphing framework leveraging StyleGAN2's latent space. The framework includes an advanced encoder, ReStyle-ID, designed to embed identities into StyleGAN2’s latent space with high accuracy, and StyleDemorpher, a specialized face demorphing network trained on the newly created DemorphDB dataset. DemorphDB features high-quality morph images, providing a challenging and realistic training environment for the StyleDemorpher. The ReStyle-ID encoder and StyleDemorpher frameworks collectively enhance the accuracy and quality of face demorphing, addressing limitations such as low resolution and poor generalizability in previous approaches. The ReStyle-ID encoder utilizes improved loss functions and training data, achieving improvements in identity reconstruction when compared to other encoding methods. StyleDemorpher excels in reconstructing high-quality demorphed images, demonstrating high generalizability across various morphing methods and unseen identities. This work introduces a robust solution for face demorphing and sets the stage for future advancements by developing a comprehensive dataset and a scalable framework for ongoing research in biometric security. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Programme: | Electrical Engineering MSc (60353) |
| Link to this item: | https://purl.utwente.nl/essays/103290 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page