[论文解读] Deep Learning for Iris Recognition: A Review

Yimin Yin, Siliang He+4 more|arXiv (Cornell University)|Mar 15, 2023

Biometric Identification and Security被引用 11

一句话总结

对2016–2022年深度学习在虹膜识别中的方法的全面综述，涵盖识别、分割、呈现攻击检测和定位，含数据集与挑战。

ABSTRACT

Iris recognition is a secure biometric technology known for its stability and privacy. With no two irises being identical and little change throughout a person's lifetime, iris recognition is considered more reliable and less susceptible to external factors than other biometric recognition methods. Unlike traditional machine learning-based iris recognition methods, deep learning technology does not rely on feature engineering and boasts excellent performance. This paper collects 120 relevant papers to summarize the development of iris recognition based on deep learning. We first introduce the background of iris recognition and the motivation and contribution of this survey. Then, we present the common datasets widely used in iris recognition. After that, we summarize the key tasks involved in the process of iris recognition based on deep learning technology, including identification, segmentation, presentation attack detection, and localization. Finally, we discuss the challenges and potential development of iris recognition. This review provides a comprehensive sight of the research of iris recognition based on deep learning.

研究动机与目标

• Motivate and frame iris recognition as a stable, secure biometric modality and assess how deep learning improves over traditional handcrafted features.
• Catalog and analyze widely used iris datasets and their characteristics to guide DL-based research.
• Summarize deep learning approaches across key iris recognition tasks (identification, segmentation, PAD, localization) and compare architectures and training strategies.
• Identify current challenges and suggest potential future research directions for DL-based iris recognition.

提出的方法

• Systematic review of 120 papers on deep learning for iris recognition published 2016–2022.
• Taxonomy of iris recognition tasks into identification, segmentation, presentation attack detection (PAD), and localization.
• Discussion of network architectures used for segmentation (FCN, U-Net) and feature extraction (AlexNet, DenseNet, ResNet, custom networks).
• Analysis of datasets (IITD, UBIRIS.v2, ND-IRIS-0405, CASIA variants, others) and their role in DL evaluation.
• Synthesis of matching/classification approaches (SVM, KNN, HD, cosine, Euclidean) and fusion of handcrafted with DL features.
• Highlighting challenges such as dataset biases, variability, and presentation attack issues, and proposing directions for future work.

实验结果

研究问题

• RQ1 What are the most influential datasets and how do they support DL-based iris recognition research?
• RQ2 What DL architectures and processing pipelines (end-to-end vs non-end-to-end) are used for iris identification, segmentation, PAD, and localization?
• RQ3 How do DL methods perform for iris identification compared to traditional handcrafted features and classical classifiers?
• RQ4 What are the main challenges and open research directions in applying DL to iris recognition?

主要发现

• DL techniques automatically learn high-level iris representations and often outperform handcrafted features in feature extraction tasks.
• DenseNet-based features achieved high accuracies, e.g., 98.7% on LG2200 (sixth layer) and 98.8% on CASIA-Iris-Thousand (fifth layer).
• On suitable datasets, CNN-based feature extraction combined with classifiers achieved strong results, e.g., 96.41% ACC on CASIA-Iris-Thousand with ResNet-50 features and PCA, and 99.4% RR on IITD in another study.
• Non-end-to-end iris identification approaches still rely on pre-processing, feature extraction, and matching steps, with SVM, HD, cosine similarity, and MLP as common classifiers/matchers.
• Several public iris datasets are widely used in DL research, including IITD, UBIRIS.v2, ND-IRIS-0405, MICHE-I, CASIA-V4 variants, Clarkson, Warsaw, Notre Dame, and IIITD-WVU.
• The survey provides a structured overview of DL-based iris recognition tasks and highlights current gaps and future directions (datasets, segmentation, PAD, localization, and end-to-end learning).