Understanding Iris Biometrics: Analysis of iris Feature patterns with CNN and Generalized Structure Tensor

Abstract

The field of iris recognition is widely studied due to its broad use in security contexts, such as airports and border control. This paper presents a descriptive study of an iris feature extraction approaches that utilize Convolutional Neural Networks (CNNs) and Generalized Structure Tensor (GST). We investigate how CNNs and GST techniques extract and analyze iris characteristics, comparing their effectiveness and potential combinations. The objective of our research is to gain a deeper understanding of the capabilities of these technologies and their potential impact on iris recognition systems. The experiments were carried out on our own generated KVK-R iris recognition dataset, with limited training images per class. The results look good; our models worked better than the previous ones. We also present a visualization method for identifying important regions in iris images that might significantly affect identification results. We think that a lot of other biometrics can be recognized using this method. CNN, on the other hand provide a robust approach to iris pattern recognition.  By training CNN model on iris dataset, it automatically learning and detecting complicated patterns within iris images. The proposed iris recognition method was tested on two public datasets: KVK-R and CASIA-Iris-Thousand. The approach produced excellent results with high accuracy rates. Using trained models for feature extraction allowed for the recognition of both closed and open sets. The best results yet to be achieved with 97% for KVK-R and 89% for CASIA-Iris thousand are achieving state-of-the-art results.