A qualitative analysis on Iris localization techniques and approaches of Machine Learning in Iris recognition systems

Abstract

Biometry is usually regarded as the most accurate technique of recognizing a person and creating a link between them and their associated identity information. Biometric authentication offers a fair level of assurance in identifying a person and providing access to approved devices with minimal discomfort. The use of a person's unique iris patterns to identify them has become a popular approach in a variety of fields.  The richness and immutability of the iris texture make it suitable for very accurate and automated identification. It effectively precludes the chance of two persons with similar iris patterns being recognized. A typical iris authentication system includes image capture, iris localization, iris normalization, feature encoding, pattern matching, and recognition decision. The most frequent stage in iris scanning, iris localization, has gained favor in the evaluation of the numerous techniques employed. It is generally understood that users of biometric systems may have varied degrees of recognition accuracy. This article examines the phases of iris identification and compares contemporary iris localization approaches in terms of iris normalization accuracy using the accessible iris datasets: IIT Delhi V1.0, MMU V1.0, and CASIA Iris Interval. It also offers a qualitative assessment of the various Machine Learning techniques used in today's Iris Recognition Systems.