AI Based Identity Verification using IRIS as a Biometric Modality

Abstract

The aim of our work is to create a sophisticated eye iris detection system that leverages the power of machine learning algorithms to achieve precise and real-time identification of the iris region in digital eye images. The system aims to offer a user-friendly interface for capturing eye images or accepting image inputs, and it will display the detected iris region with relevant information. The ultimate goal is to develop a versatile solution that can be integrated into biometric authentication or security systems, enabling effective identity verification and facilitating various applications in surveillance, healthcare, and other domains by using CNN (Convolutional Neural Network) and RSNN (Reservoir Computing-based Spiking Neural Network.

Introduction

I. INTRODUCTION

In an era of heightened security and privacy concerns, the demand for robust identity verification systems has surged. Traditional authentication methods like passwords and PINs are vulnerable to various attacks, making biometric modalities an attractive solution. Among these, iris recognition shines as a remarkably secure and accurate means of identity verification. The intricate and stable iris patterns, formed during fetal development and enduring throughout life, provide unmatched uniqueness in biometrics. The human iris, with its distinctive features, is recognized as a highly reliable biometric marker. This project aims to harness the potential of the iris by developing an advanced machine learning-based eye iris detection system, enabling precise identification and localization of the iris region in digital eye images. Leveraging the iris's reliability and immutability, this system promises heightened security and efficiency for authentication and identification in diverse applications.

II. LITERATURE SURVEY

A deep learning-based iris recognition system [1] called Deep Iris Net. The authors employ a 1D CNN architecture to extract discriminative features from the iris images. They also introduce a large-scale iris dataset, and their experiments demonstrate that Deep Iris Net achieves competitive performance compared to traditional iris recognition methods. An efficient iris recognition system using a convolutional autoencoder. The proposed method leverages the autoencoder to learn compact iris representations. The extracted features are then used for classification tasks. The authors demonstrate [2] the effectiveness of their approach on publicly available iris datasets, achieving promising recognition rates. In the work [3], the authors propose a multiscale CNN architecture for iris recognition. The model operates at multiple scales to capture both fine-grained and coarse-grained iris patterns. The proposed approach is evaluated on standard iris datasets, and the results show improved recognition accuracy compared to conventional methods. A two-stream CNN architecture for iris recognition [4]. The authors employ two parallel CNN streams—one processes the iris texture, and the other processes the iris. contour. The features extracted from both streams are fused to improve recognition performance. Experimental results on benchmark iris datasets demonstrate the effectiveness of the proposed approach. A multitask learning approach for iris recognition using CNNs. The model [5] simultaneously learns to perform iris segmentation and recognition tasks. The multitask learning framework enables the network to share knowledge between tasks, leading to better generalization and improved recognition accuracy.

III. PROPOSED SYSTEM

The proposed system introduces an innovative solution that surpasses the constraints of existing methods by utilizing a hybrid CNN-RNN algorithm. It comprises several key components starting with data collection and preprocessing, involving tasks like resizing, normalization, and data augmentation to enhance data quality. Feature extraction is accomplished using a CNN architecture, which excels at capturing intricate patterns in iris images. Subsequently, extracted features are transformed into sequences and processed through RNN layers to capture temporal dependencies.

The hybrid model architecture seamlessly integrates the strengths of both CNNs and RNNs, with CNN layers focusing on feature extraction and RNN layers capturing sequence information. Training and optimization involve using a curated dataset and tuning hyperparameters for optimal performance. Model evaluation encompasses metrics like accuracy, precision, recall, and F1-score, with fine-tuning based on evaluation results. Finally, the optimized hybrid model undergoes testing on a separate dataset to assess its generalization capability, potentially paving the way for real-world iris-based person identification tasks. This holistic approach capitalizes on the synergistic relationship between CNNs and RNNs, aiming to significantly enhance accuracy and reliability in iris-based person identification.

IV.  METHODOLOGY

The methodology comprises several steps. Initially, it involves gathering a diverse dataset of labelled eye images that include both iris and non-iris regions for training and evaluation purposes. Following this, the collected data undergoes preprocessing to eliminate noise, normalize intensity, and standardize resolution. Subsequently, relevant features are extracted from the pre-processed images to effectively represent the iris region and same is shown in the below figures 1, 2, and 3.

VI. ACKNOLEDGEMENT

We would like to express our sincere thanks and indebtedness to my esteemed institution, The National Institute of Engineering, Mysuru which has provided me with an opportunity to fulfil my desire and reach my goal. Their insightful feedback and encouragement at every stage of the project have been truly inspiring. We extend our heartfelt appreciation to our family and friends for their unwavering love, encouragement, and understanding throughout our academic journey.

Conclusion

In summary, iris detection stands as an innovative and effective system that harnesses the power of machine learning. Through a thorough evaluation process, we assessed multiple models including CNN, RSNN this underscores its ability to capture. Our meticulous analysis extended to precision and recall values across distinct iris patterns categories, offering valuable insights into model performance for each. weighted averages allowed us to comprehensively gauge model effectiveness across categories. Training of dataset is done with high priority so as to pull the maximum accuracy possible by the system mainly focusing on the required pattern.

References

[1] Python ML, Third Edition, Sebastian Raschaka, Vahid Mirjalili [2] A hybrid approach to building iris retina detection recommender system. Kitsuchart pasupa, wisuwat sunhem, chu kiong loo [3] Deep iris .net: A new large scale data set for iris recommendation based on yutao chen, Yuxuan zhang [4] Python ML, Third Edition , Sebastian Raschaka, Vahid Mirjalili [5] A hybrid approach to building iris retina detection recommender system. Kitsuchart pasupa, wisuwat sunhem , chu kiong loo [6] Deep iris .net: A new large scale data set for iris recommendation based on yutao chen, Yuxuan zhang

Copyright

Copyright © 2023 Dhanush R, K R Sumana . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.