A novel approach for iris recognition

Abstract

Iris is one of the biometrics that is very stable and reliable. This research aims to develop a method that can be used to perform iris identification using four hundred and twenty iris images from CASIA-v4-Syn (non-ideal iris) along with three hundred and seventy eight iris images from CASIA-v1 (ideal iris). The basic principle of the proposed method as follows: firstly, design the preprocess using amoeba median filter and Gaussian filter in order to enhance the effective area of the iris. Secondly, using modified CHT method, the iris area is segmented to separate the iris from the pupil. Subsequently, the selection process of ROI by applying homogeneous rubber sheet model to extract the intersection of internal and lower region from the segmented iris will consequently transform it to a rectangular shape which will be used for features extraction. The characteristic features namely maximum probability, correlation, contrast, energy, homogeneity, and entropy are then extracted using multiple 3D-GLCM; which is the advanced version of 2D GLCM. Finally, these features are trained using Elman Recurrent Neural Network/Levenberg-Marquardt algorithm to obtain the accuracy. Studies have shown that the recognition rates of this method can reach CRR of 91% using CASIA-Iris-Syn v4 and CRR of 94.22% using CASIA-Iris-Syn v1 which can meet the demand of iris recognition.