Distant iris recognition through machine learning models with deep features transfer for human identification

Human identification through biometrics has become increasingly popular due to its reliable authentication in automated high-security surveillance systems. Several biometric models based on fingerprint, face detection, and iris recognition have been designed and developed for human identification. Among these biometrics, iris recognition, especially distance-based recognition, remains a significant challenge due to its small imaging target. In this paper, we propose a distant iris-based human identification framework employing a deep extracted feature transfer with machine learning (ML) models. In the first stage, we customized the traditional convolutional neural network (CNN) model and utilized three pre-trained models VGG16, VGG19, and ResNet50 for the extraction of deep features from normalized iris images. Later, we fed these deep features extraction into nine ML models for iris image classification. The proposed framework is validated via several experiments using the CASIA-V4 iris dataset. Experimental results show that the softmax classifier with our customized CNN model outperforms the considered pre-trained deep learning models, achieving top scores in accuracy (93.40%), precision (94.31%), recall (93.40%), F1-score (93.25%), and Cohen’s kappa (93.34%). This customized CNN model with a softmax also demonstrates competitive performance when compared with other distance-based iris recognition models.