IMPROVING IRIS RECOGNITION ACCURACY USING GABOR KERNELS WITH NEAR-HORIZONTAL ORIENTATIONS

Abstract

extraction of iris features. However, the selection of Gabor kernel orientations remains an important key for an optimum performance. In this paper, Gabor kernels with horizontal and two near-horizontal orientations (0°, 15°, 165°) are used for extracting iris features, as an attempt to improve the performance of iris recognition rate. The three iris features are cascaded into one image for the purpose of image matching. To this end, an iris recognition system is developed using methods of iris localization, eyelid removal and iris matching by Hamming Distance (HD) that have been developed previously by the same authors. The system is implemented on three standard datasets, CASIA-1, CASIA-Lamp-4 and SDUMLA-HMT. The results have shown that the overall accuracy (Accu), the True Positive Rate (TPR) and the Equal Error Rate (EER) EER are (98.85%, 99.42% and 0.58%) for CASIA-1 database, (96.56%, 98.25% and 1.75%) for CASIA-Lamp-4 database and (96%, 98% and 2.0%) for SDUMLA-HMT database. These results outrage the results of most of the previous works that have used the same databases.