A Systematic Review of Deep Learning-Based Iris Recognition: Methods, Datasets, and Performance Evaluation

Abstract

Biometric systems are pattern recognition technologies used to identify or verify individuals by collecting their biometric characteristics, extracting features that can be compared with stored templates in a database. Among the different biometric modalities, iris recognition has attracted particular interest due to the presence of highly distinctive patterns in the iris itself that is unique to every individual and those patterns will NEVER change. This new paradigm is shown by the success of deep learning, particularly Convolutional Neural Networks (CNN), which are capable to learn descriptive representations with complex image data and significantly boost on the performance of iris recognition. This review will discuss 16 related studies published between 2020 and 2026 that give us a wide view of iris recognition methods based on deep learning. This paper defines the main components of the iris recognition pipeline such as image pre-processing that aims at enhancing the quality of iris images, iris segmentation where first segments an isolated region containing only the iris and other features from a generic background, feature extraction where deep learning models should be able to produce effective representation learnt for irises and lastly classification using extracted features as input for identification or verification tasks. Additionally, the review provides a summary of the most widely used publicly available datasets in iris recognition studies and compares the performances reported by all selected works with respect to computational efficiency and accuracy.