Employing spectral signature for ocular eyeglass detection

Abstract

User verification performance based on ocular region has been greatly affected when user wear eyeglasses. Therefore, to improve the verification accuracy of ocular biometric system, the eyeglass detection has been explored in recent times. In this work, we present the ocular glass detection by using multi-spectral imaging approach in eight narrow spectrum bands across Visible and Near-Infra-Red wavelength range. Specifically, this work is based on extracting spectral signature and then performing classification by learning the features through linear Support Vector Machine (SVM) classifier model. The approach is performed on 16640 ocular instances to presents the significance of this work. We compare the performance evaluation across six different feature extraction methods. We repeated the experiment 10 times by selecting training and testing data samples randomly for each trial to present the average value of Equal Error Rate to detect ocular eyeglass. The lowest EER value of 12.04 percent is obtained using Spectral Signature based approach demonstrating the effectiveness of employing multi-spectral imaging for eyeglass detection in ocular biometric.