Method for automatic pupil segmentation in ophthalmological images with anomalies

The purpose of the research is to develop a technique for segmentation of pupils with pathological changes in shape, overcoming the limitations of traditional methods based on circle search. The key task is to overcome the fundamental limitation of traditional algorithms, which assume that the pupil has a shape close to a circle and are untenable when dealing with anomalies such as synechia, leading to oval, star-shaped and other complex contours.

Methods. The paper proposes a technique for automated pupil segmentation, which includes sequential image processing using the following key steps: pre-filtering (FIR filtering), illumination correction, logarithmic and power trans formations, morphological processing and removal of artifacts. To segment pathologically altered pupils, a combination of adaptive algorithms was used, which makes it possible to effectively process non-circular pupils (oval, star shaped, etc.), which traditional methods based on circle search (Hough transform, DIDO, etc.) cannot cope with.

Results. The experimental evaluation was carried out on a specialized database of ophthalmological images with various deformations. The proposed technique demonstrated segmentation accuracy exceeding 90% for pupils with deformations, which is significantly higher than the results shown by classical methods such as the Hough transform and DIDO. Detailed analysis showed that 92.3% accuracy was achieved for oval–shaped pupils, 89.7% for the most difficult cases of star‒shaped pupils, confirming the efficiency of the technique.

Conclusion. The developed technique has proven its effectiveness for reliable segmentation of pathologically altered pupils, which standard approaches cannot cope with. High accuracy opens up opportunities for its practical application in two key areas: in highly reliable biometric identification systems, where pupil deformities are natural noise, and in medical diagnostic support systems for the quantification of pathologies. A promising area of further research is the adaptation of the methodological approach for segmentation of other ophthalmological structures with irregular boundaries, such as the pterygium

1. Bowyer K.W., et al. Image understanding for iris biometrics: A survey. Computer Vision and Image Understanding. 2018;167:1-21.

2. Raghavendra R., et al. A Comprehensive Survey on Iris Recognition Systems: A Review. IEEE Access. 2021;9:45615-45630.

3. Komkova S.V. A method for diagnosing glaucoma from human fundus images. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Serija: Upravlenie, vychislitel'naja tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the Southwest State University. Series: Control, Computer Engineering, Information Science. Medical Instruments Engineering. 2023;13(3):99‒114. (In Russ.)

4. Komkova S.V. Methodology for forming a feature vector based on retinal images. Estestvennye i tekhnicheskie nauki = Natural and Technical Sciences. 2021;(4):250-252. (In Russ.)

5. Komkova S.V. Algorithm for identification of exudates in human retinal images // Vestnik komp'yuternykh i informatsionnykh tekhnologii = Bulletin of Computer and Information Technologies. 2022;19(1):47-51. (In Russ.) https://doi.org/10.14489/vkit.2022.01. pp.047-051

6. Komkova S.V. "Methodology for detecting hard exudates in human fundus images" // Telekommunikatsii = Telecommunications. 2022;(10):24-28. (In Russ.)

7. Chen Y., et al. An Adaptive Algorithm for the Detection of Irregular Pupil Contour. Journal of Medical Systems. 2019;43(5):112.

8. Radman A., et al. Automated Pupil Segmentation and Noise Removal in Iris Images. Journal of Medical Imaging and Health Informatics. 2017;7(2):428-436.

9. Wang C., Muhammad J. Iris segmentation and recognition using deep learning. Neural Computing and Applications. 2020;(32):14579-14590.

10. Uhl A., Wild P. Weighted Adaptive Hough and Ellipsopolar Transforms for Realtime Iris Segmentation. In: Proceedings of the 5th International Conference on Biometrics (ICB’12). New Delhi, India; 2012. P. 283–290. https://doi.org/10.1109/ICB.2012.6199821

11. Zhao Z., Kumar A., et al. An Accurate Iris Segmentation Framework Under Relaxed Imaging Constraints Using Total Variation Model. In: IEEE International Conference on Computer Vision (ICCV). Santiago, Chile; 2025. https://doi.org/10.1109/ICCV.2015.436

12. Hofbauer H., Jalilian E., Uhl A. Exploiting Superior CNN-based Iris Segmentation for Better Recognition Accuracy. Pattern Recognition Letters. 2019;120:17–23. https://doi.org/10.1016/j.patrec.2018.12.021

13. Wang K., Kumar A. Toward more accurate and constrained iris segmentation using cascaded multi-task framework. IEEE Transactions on Information Forensics and Security. 2021;(16):2023-2036.

14. Ivanov A.V., Petrov S.K. Iris segmentation algorithms for pathological pupils. Biomeditsinskaya radioelektronika = Biomedical Radioelectronics. 2020;(3):45-53. (In Russ.)

15. Smirnova E.A., et al. Methods for processing ophthalmic images with anomalies. Opticheskii zhurnal = Optical Journal. 2021;88(5):78-85. (In Russ.)

16. Kozlov V.P., Nikolaeva T.G. Automated analysis of fundus images. Komp'yuternaya optika = Computer Optics. 2019;43(4):623-630. (In Russ.)

17. Gusev D.A. Digital processing of biometric images. Moscow: Tekhnosfera; 2018. 320 p. (In Russ.)

18. Belov P.N., et al. Iris recognition algorithms for eye pathologies. Iskusstvennyi intellekt i prinyatie reshenii = Artificial Intelligence and Decision Making. 2020;(2):34-42. (In Russ.)