Information and communication and chemical technologies

Vol. 3 No. 28 (2025) / 2025-09-30 / Number of views: 176

LITERATURE REVIEW AND COMPARATIVE ANALYSIS OF IRIS IDENTIFICATION METHODS

Authors

L.N.Gumilyov ENU
https://orcid.org/0000-0003-3795-0424
L.N.Gumilyov ENU
https://orcid.org/0000-0002-8405-5038
L.N.Gumilyov ENU
https://orcid.org/0000-0002-9162-6791
L.N.Gumilyov ENU
https://orcid.org/0000-0002-8152-8661
Esil University
https://orcid.org/0000-0002-8435-7773

Keywords

iris, authentication, deep learning, image segmentation, challenges in open environments

Link to DOI:

https://doi.org/10.58805/kazutb.v.3.28-874

How to quote

Mussaif, M., et al. “LITERATURE REVIEW AND COMPARATIVE ANALYSIS OF IRIS IDENTIFICATION METHODS”. Vestnik KazUTB, vol. 3, no. 28, Sept. 2025, doi:10.58805/kazutb.v.3.28-874.

Abstract

Throughout our work, we relied on open databases of images of the iris. CASIA and UBIRIS proved to be particularly useful among them. They include images taken in a variety of conditions, from good laboratory lighting to more "real-life" situations. This is exactly what is needed to test how well the algorithms are coping outside of an ideal environment.

Before the analysis, the photos were slightly prepared: contrast was improved, noise was removed, brightness was adjusted. All this was done to ensure that the structure of the iris was clear and suitable for further processing.

Next came the stage where it was important to find and highlight the iris itself. The Hough method has proven itself well here, as it knows how to accurately find circles. After that, the shape of the iris was transformed from circular to rectangular, so that it would be easier to work with the images in the following steps.

When the form is sorted out, it's time to extract the signs. These are such unique features that you can tell one eye from the other. We used Gabor filters and wavelets -they do a good job with detailed texture. Then the classification: the system must understand whose eye it is. Neural networks and the support vector method helped here - together they gave fairly accurate results.

During the testing process, we paid attention not only to the accuracy, but also to the speed of operation. It is also important to keep track of when the system makes mistakes, both in one direction and in the other. Such failures can affect security.

The final step is statistics. We compared different versions and tracked which ones performed better. This is how we managed to understand where the algorithm is weaker and where it works stably. And this makes it possible to move on and improve the system.