Iris Recognition via Deep Learning Using Capsule Networks with Enhanced Routing Algorithm

Document Type : Original Article

Authors

1 Qom University of Technology

2 University of Genoa

Abstract
Iris recognition is a widely used biometric technology in various applications. Deep learning methods, particularly convolutional neural networks (CNNs), have been popular in biometric detection due to their ability to generalize well and operate without human intervention. However, CNNs often struggle with image noise and require large datasets for effective training. To address these limitations, Capsule Networks (CapsuleNet) have been introduced, offering improved performance on small and noisy datasets. This paper presents a CapsuleNet architecture with an enhanced routing algorithm tailored for iris recognition. To further refine the learning process, VGG16 and InceptionV3 models are integrated into the CapsuleNet, enabling it to learn effectively from a limited number of samples. The proposed network is structured into a series of subnets, corresponding to its main building blocks. Experiments conducted on the CASIA-V4 Lamp iris dataset demonstrate that the optimized CapsuleNet architecture outperforms traditional networks, offering greater stability and robustness for iris recognitions. Results on the CASIA-V4 dataset show a 6% improvement in accuracy.

Keywords

Subjects

Volume 1, Issue 2
Spring 2025
Pages 91-99

  • Receive Date 08 October 2024
  • Revise Date 16 November 2024
  • Accept Date 18 November 2024
  • First Publish Date 01 April 2025