Coarse iris classification by learned visual dictionary

Abstract

In state-of-the-art iris recognition systems, the input iris image has to be compared with a large number of templates in database. When the scale of iris database increases, they are much less efficient and accurate. In this paper, we propose a novel iris classification method to attack this problem in iris recognition systems. Firstly, we learned a small finite dictionary of visual words (clusters in the feature space), which are called Iris-Textons, to represent visual primitives of iris images. Then the Iris-Texton histograms are used to represent the global features of iris textures. Finally, K-means algorithm is used for classifying iris images into five categories. Based on the proposed method, the correct classification rate is 95% in a five-category iris database. By combining this method with traditional iris recognition algorithm, our system shows better performance in terms of both speed and accuracy. © Springer-Verlag Berlin Heidelberg 2007.