Race Classification Based Iris Image Segmentation

Abstract

Iris segmentation is an essential precondition for biometric authentication systems based on iris recognition and dramatically affects the accuracy of personal identification. Due to various noises during iris acquisition, iris images from different databases exhibit different texture characteristics. Existing works mostly design segmentation schemes for specific iris images and thus restrain much room for performance improvement. Therefore, this paper proposes a race classification based iris image segmentation method. Compared with conventional methods, the proposed method firstly exploits the merits of local Gabor binary pattern (LGBP) with support vector machine (SVM) and builds an efficient classifier, LGBP-SVM, to partition iris images into the human eye and non-human eye images. Following this, these two kinds of iris images are segmented by different strategies based on circular Hough transform with the active contour model. Extensive experiments demonstrate the proposed LGBP-SVM outperforms existing works in terms of accuracy of iris race classification. Furthermore, the race classification based iris segmentation method improves the segmentation accuracy and correct segmentation rates for various iris image databases.