Improved Iris-SAM -Based Iris Segmentation for Recognition in Biometric Security
Maduabuchi Kingsley Okorie, Ndubuisi John Ngene, Uchechukwu Chioma Ugwa, Gabriel Oko Ota · European Journal of Applied Science Engineering and Technology · 2026
Accurate iris segmentation is a critical challenge in biometric domain as a result of occlusions, illumination variations, and noise in unconstrained environments. To tackle these challenges, this research adopted Improved Iris-SAM built on IrisAdapter framework to optimize the accuracy of iris segmentation. The framework was built on IrisAdapter and leveraged the effectiveness of SAM. The dataset used for the experiment included CASIA-Iris-Interval-v3, ND-Iris-0405 and IIT-Delhi-Iris. The result showed that the proposed Improved Iris-SAM framework achieved high iris segmentation performance with average IoU scores of 97.65%, 99.86% and 95.53% respectively. The outcomes outperformed traditional and deep learning-based benchmark methods such as OSIRIS and SegNet. The proposed framework indicated strong generalization ability across-dataset evaluations. However, the Improved Iris-SAM framework extensively improved robustness and accuracy in iris segmentation and enhanced overall biometric recognition performance.