Real time iris segmentation quality evaluation using medoids. (May 2023)

• Record Type: Journal Article
• Authors: Ejiogu, Ugochi U.C.; Iloanusi, Ogechukwu N.
• Is Part Of: Pattern recognition. Volume 137(2023)
• Publication Date: 2023-05
• Subjects: Iris segmentation-quality estimation -- Biometric vision and computing iris dataset -- K-medoids clustering -- Iris datasets -- Iris ground-truth mask -- Eccentricity -- Annulus radii ratio -- Euclidian distance -- Iris mask -- medoids; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.109290
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose a novel medoids based iris segmentation-quality evaluation model, as a robust alternative to real-time iris segmentation-quality evaluation. It uses eccentricity measure and annulus radii ratio to screen out severe segmentation failure prior to k-medoids clustering. A comparative evaluation using CASIA_v4, IITD_v2, UBIRIS_v2 and a novel iris dataset Biometric Vision and computing (BVC_v1s1) iris dataset was performed. The proposed model consistently recorded the least classification error-rate across diverse iris datasets and 95.27% classification accuracy rate on one evaluating iris-datasets. Abstract: Integrating an efficient and robust iris segmentation-quality estimation module in iris biometric systems will undoubtedly enhance its performance and competitive advantage. It proffers a real time detection of segmentation errors to forestall their propagation to the subsequent modules. Hence, we propose a novel automatic iris segmentation-quality estimation model using medoids. The performance of the proposed model was empirically evaluated with reference to three published models, using three benchmarked iris datasets and our novel proprietary iris dataset – Biometrics Vision and Computing Iris dataset. The proposed medoids based model was experimentally demonstrated to be effective, robust and relatively efficient in estimating iris segmentation-quality. Specifically, the proposed model recorded the best classification accuracy rate of 95.27% on one of the datasets. Also, it consistently recorded the least classification error rate across several iris datasets with diverse segmentation-errors, which suggest that the medoids based model is relatively more robust than the examined counterparts.