Feature Extraction Method based on Various Scanning Techniques in Iris Recognition System

Abstract

Reliable personal recognition is demand of time due to increase of terrorism and criminal offences. So, modern societies give more importance to the systems that contribute to increase of security and reliability. One of the techniques for this purpose which has proven to be very accurate and reliable is iris recognition. Iris recognition is basically done in five steps namely; iris image acquisition, iris segmentation, iris normalization, feature extraction and feature matching. For good performance of Iris Recognition System, segmentation and normalization play an important role. Also to extract the required features from the normalized iris, feature extraction techniques are adopted. In this thesis, methods of localization and feature extraction based on various scanning techniques in iris recognition are presented. The segmentation process in Iris Recognition System is performed based on Daugman’s Integro-differential operator which is capable of localizing the iris region by assuming iris and pupil as perfect circles. Localized iris is then normalized based on Daugman’s Rubbersheet model. After applying Discrete Cosine Transform on the normalized iris a method for iris feature extraction using various scanning techniques on the obtained Feature Vector coefficients is proposed. The scanning techniques used are Zigzag, Raster, and Sawtooth. The previous work for iris feature extraction using Discrete Cosine Transform provides good results but approach of using scanning techniques in this thesis improves the accuracy percentage. Experimental results show the promising performance of Raster Type-II scanning technique with 80.30% accuracy when 100 coefficients are taken. The database used for the observations is CASIA iris database version-IV.