Design and Analysis of Deep-Learning based Iris Recognition System

Abstract

Biometrics The existing iris identification techniques that have been published over the years primarily rely on particular circumstances, such as the distance at which the images are acquired and the setting of constant staring, which necessitates extensive user collaboration. These segment/normalize based and "phase based" method encountersnumerous issues when representing data reduces the disparities caused by heavy eyelash occlusion, motion blurs, motion translations, scale, rotations, pupillary dilation, and irregular reflections in the area around the user's eyes in situations where there is no assurance of collaboration. By analyzing the displacements between the respective patches in pairs of iris pictures, convolutional neural network-based deep learning classification models (CNN) are able to accurately distinguish between genuine and imposter comparisons using iris segmentation, even in extremely noisy environments. When compared to other biometric recognition systems, the iris texture's ability to differ across identical twins' eyes and between the left and right eyes of the same person gives it a more secure method of authentication. In the studies, we took into account three well-known data sets (CASIA, Polaris), coming to the conclusion that the suggested algorithm is effective, especially in situations when precisely segmenting the iris is difficult.