Semester of Gradation
Summer 2025
Degree Type
Dissertation
Degree Name
Doctor of Philosophy in Data Science and Analytics
Department
Data Science and Analytics
Committee Chair/First Advisor
Mahmut Karakaya
Second Advisor
Ramazan Aygun
Third Advisor
Md Abdullah Al Hafz Khan
Abstract
The usage of biometrics offers a reliable, and high performing technology in the feld of person identifcation and verifcation. Iris recognition, which uses the human iris as the biometric, is known for its low False Acceptance Rates and False Rejection Rates, making it one of the most secure and accurate forms of biometric authentication available today. However, most of the current iris recognition systems are based on frontal iris recognition, which limits the usability of the technology. In frontal iris recognition, it is required that the axis of the human eye aligns with the axis of camera lens. To make iris recognition unconstrained and expand on its usability, there is a need for off-angle iris recognition. In off-angle iris recognition, there is no need for the axis match of human eye and camera lens, making off-angle iris recognition suitable for tasks like fast security checks and remote surveillance. However, off-angle iris recognition is not as simple as performing a perspective projection, because of several factors such as illumination, presence of cornea and limbus in the human eye. In this thesis, we propose three models that aim at improving the off-angle iris recognition at different levels. The frst model identifes the most affected regions in the iris images, and masks them. The second model creates a frontal projection of the off-angle iris image using a modifed Pix2Pix GAN model. The third model proposes a novel segmentation algorithm to improve the performance to automate and improve the off-angle iris recognition performance.