Implementation of a Customized Light CNN Architecture for Iris Recognition System

Biometric recognition refers to a technological approach that enables the identification and authentication of individuals by leveraging their distinct physical or behavioral attributes. The utilization of this technology is prevalent in the domains of security, access control, and authentication of identity. Unlike traditional identification methods such as passwords or PINs, biometric recognition relies on the distinctive traits of an individual, making it more secure and difficult to fake. Commonly used biometric approaches include fingerprint, iris, facial, palmprint, and retina. Among them, iris recognition is used widely and has unique patterns in the colored part of the eye (iris) to identify and authenticate individuals. The iris is the ring-shaped part of the eye surrounding the pupil and is known for its distinctive and stable characteristics. Even identical twins have different iris patterns, making iris recognition highly accurate and secure. However, the security aspects of the system are still unexplored. Therefore, we proposed a convolutional neural network (CNN) architecture-based approach to identify fake iris images. The suggested model includes preprocessing, saliency detection, feature extraction, and classification by CNN. All these experiments were carried out on the CASIA-Iris-Interval and CASIA-Iris-Syn databases. Through this process, the implemented technique attained 100% accuracy.