Texture Classification in Iris Using Dominant Neighborhood Structure
J. Senthil Raju, C. Anand, Deva Durai · 2013
Abstract— This paper proposes an approach to extract global image features of iris for the purpose of texture classification. Iris is needed to be segmented to get the textures. Hough transform is used to segment the iris region and the segmented region is normalized by using Daugman's rubber Sheet Model. Features can be obtained by using the local binary patterns (LBPs) and dominant neighborhood structure (DNS) methods. LBP are used to extracted local texture features and DNS for global features. DNS, is estimated by measuring the intensity similarity of a given image pixel to its surrounding pixels within a certain local image neighborhood called search window, that is not only rotation-invariant but also highly robust to noise. Both extracted features are fused together by simple sum fusion method which is complement to each other, so increase the classification accuracy. Then classification can be performed by using SVM classifier. CASIA iris database is used in this paper.