A Novel Approach for Iris Localization using Machine Learning Algorithms

Iris recognition is a proven and highly reliable method for biometric security applications due to the uniqueness of texture of each individual's iris. While this may be the case, the process of localization of human iris is challenging due to eyelids and eyelashes, reflections and blurring acting as noise for the localization process and are present in the normalized iris outcome as well. This paper presents a fast and reliable method of eyelid noise reduction using Daugman's rubber sheet model. Pre-processing methods have been applied to reduce noise and to achieve sharp edge detection for application of circular Hough transform. For normalization, Daugman's rubber sheet model is used on which selective angular segmentation along with radii reduction and cropping is performed to achieve a clear iris band. On this segmented iris band, a hybridized feature vector creation technique involving calculation of grey level co-occurrence matrix along with wavelet decomposition has been applied for creation of feature vectors which were passed to a list of machine learning classifiers for performance evaluation.