Improved segmentation algorithm and further optimization for iris recognition

Abstract

Iris recognition is a biometric authentication system proving vital for ensuring security and has been employed as an important case to test the algorithms developed in pattern recognition. The unique circular shape of the iris and its time invariance makes it a versatile technique that has an accuracy that can be mathematically proven. Here in this work we propose a new segmentation technique and two new encoding schemes. The newly proposed techniques are tested on the best, the worst and on all the irises of two widely known databases (CASIA and IIT Delhi database) and the results are compared with the classical segmentation and classical encoding schemes. The segmentation is improved and as a result the accuracy and equal error rate also. In CASIA database the use of the new segmentation improves the EER from 3.14% to 0.82% (on all 756 images of the dataset). When tested on the whole IIT iris database, the EER is improved from 3.88% to 0.34%; and on the worst images of IIT EER is improved from 13.30% to 1.00%.