An Efficient Automated Iris Recognition Classification Model Based on Different Convolutional Neural Networks and Transfer Learning.

Authors

Ghada Abd El-Latif shatat, Department of Communication & Electronics, Nile Higher Institute for Engineering & Technology, Mansoura, EgyptFollow
Abeer Twakol, Department of Communication & Electronics, Faculty of Engineering, Mansoura University, Mansoura, EgyptFollow
Ibrahim Yasser, Communications Engineering Dept., Faculty of Engineering, Mansoura University, EgyptFollow
Mohy Eldin Ahmed Abu Al Saud, Department of Communication & Electronics, Faculty of Engineering, Mansoura University, Mansoura, EgyptFollow

Corresponding Author

Ghada Abd El-Latif

Subject Area

Electronics and Communication Engineering

Article Type

Original Study

Abstract

Iris is a common biometric used for identity verification. Iris identification is one of the best ways to give people individual authentication based on their iris anatomy. This paper's main objective is to evaluate how well these deep learning networks perform on iris image datasets. The image goes through the following stages: improving image quality; employing the Hough transform and the integro-differential operator to segment the iris; and reducing processing time by changing the image's 150x300 dimensions from Cartesian to polar coordinates. Transfer learning is used to implement the iris classification on three deep learning networks: VGG19, InceptionV3, and Iris Net. The recommended study presents several parameters, including the accuracy of each deep learning network, that were used to create an effective automated iris recognition classification model. For the iris recognition challenge, the study also compares the system's identification ability with several CNN models to determine the optimum outcome. The proposed iris recognition system is tested using Utiris-V1, CASIA Iris Twins-V3, and CASIA-Iris-V3 Interval. The system produced excellent outcomes with a high accuracy rate. Results of the proposed system show that Vgg-19 performs best, with an overall database accuracy of 1.0 and a per-person recognition time of under one second.

Keywords

Iris recognition, deep learning, Iris Net, Vgg-19 and Inception-V3, CNN.

Recommended Citation

shatat, Ghada Abd El-Latif; Twakol, Abeer; Yasser, Ibrahim; and Abu Al Saud, Mohy Eldin Ahmed (2022) "An Efficient Automated Iris Recognition Classification Model Based on Different Convolutional Neural Networks and Transfer Learning.," Mansoura Engineering Journal: Vol. 47 : Iss. 6 , Article 11.
Available at: https://doi.org/10.58491/2735-4202.3171