A robust Iris recognition method on adverse conditions

TL;DR

This paper presents a robust iris recognition method that enhances accuracy under adverse conditions by applying masking, boundary detection, wavelet-based feature extraction, and feature selection, achieving high localization and recognition accuracy on CASIA datasets.

Contribution

The study introduces a novel iris recognition approach combining masking, boundary detection, wavelet feature extraction, and feature selection for improved robustness and accuracy.

Findings

Localization accuracy: 99.73% on CASIA-v1

Feature extraction accuracy: 97.82% on CASIA3

Overall recognition accuracy exceeds 97%

Abstract

As a stable biometric system, iris has recently attracted great attention among the researchers. However, research is still needed to provide appropriate solutions to ensure the resistance of the system against error factors. The present study has tried to apply a mask to the image so that the unexpected factors affecting the location of the iris can be removed. So, pupil localization will be faster and robust. Then to locate the exact location of the iris, a simple stage of boundary displacement due to the Canny edge detector has been applied. Then, with searching left and right IRIS edge point, outer radios of IRIS will be detect. Through the process of extracting the iris features, it has been sought to obtain the distinctive iris texture features by using a discrete stationary wavelets transform 2-D (DSWT2). Using DSWT2 tool and symlet 4 wavelet, distinctive features are extracted. To reduce the computational cost, the features obtained from the application of the wavelet have been investigated and a feature selection procedure, using similarity criteria, has been implemented. Finally, the iris matching has been performed using a semi-correlation criterion. The accuracy of the proposed method for localization on CASIA-v1, CASIA-v3 is 99.73%, 98.24% and 97.04%, respectively. The accuracy of the feature extraction proposed method for CASIA3 iris images database is 97.82%, which confirms the efficiency of the proposed method.