Deep Learning for Robust Iris Recognition: Introducing Synchronized Spatiotemporal Linear Discriminant Model-Iris

Abstract

A novel Synchronized Spatiotemporal Linear Discriminant Model –Iris (SSLDMNet-Iris,) a deep learning architecture is introduced in this work which is designed to address the challenges associated with iris recognition under varying environments, such as occlusion, variations in eye pupil dilation, and lower image quality. This has been implemented by integrating multi-scale convolutional feature extraction with synchronized temporal modeling through Gated Recurrent Units (GRUs), the proposed SSLDMNet-Iris model effectively can catch both intricate texture details and global spatial patterns related to the iris. Additionally, the model utilizes Fisher’s Linear Discriminant (FLD) for features extraction and optimizing the separation between classes while minimizing intra-class variance, thereby raising recognition accuracy. Comprehensive experiments conducted on seven benchmark datasets (i.e., CASIA Iris 1.0, CASIA Iris 2.0, CASIA Iris 3.0, CASIA Iris 4.0, IITD, UBIRIS, MMU), and exhibit a promising accuracy rate where, the SSLDMNet-Iris surpassing traditional models like VGG16, AlexNet, and ResNet. Notably, SSLDMNet-Iris attains 100% accuracy on CASIA Iris 1.0, CASIA Iris 2.0, and MMU datasets, while maintaining high computational efficiency with a reduced processing time. These results highlight the robustness and versatil-ity of SSLDMNet-Iris, making it an ideal candidate for real-time iris recognition applications.