Iris biometric template identification and recognition scheme using a novel parallel fused encoder
R. Vasuki,K. Ranjith Singh
DOI: https://doi.org/10.1007/s11042-024-20415-4
IF: 2.577
2024-11-08
Multimedia Tools and Applications
Abstract:An iris recognition system is considered as an effective biometric model for human authentication. This study proposes an efficient algorithm for recognizing iris regions using a parallel fused encoder. The eye images are first pre-processed using a median filter to remove unwanted noises. Then, Daugman's algorithm is applied to these pre-processed images to localize and segment the iris portion alone. A novel parallel fused encoder technique is applied over these segmented iris portions to extract unique textural features. The output features are converted into a single vector and processed using sigmoid activation to identify whether the iris images are matched with the dataset or not. The predicted labels from this technique are compared with the original labels for performing an evaluation using Accuracy, F1-Score, Precision, and Recall. The proposed model has been experimented on three different datasets: the Multimedia University Iris Dataset (MMU), the Indian Institute of Technology Dataset (IITD), and the CASIA-IrisV4-Interval datasets. This proposed novel biometric technique showed desirable recognition accuracies of 96.00%, 93.00%, and 98.50% on MMU, IITD, and CASIA-IrisV4-Interval datasets, respectively. The obtained performances are comparatively greater than the existing classifiers used in the published literature.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering