Iris Recognition Using Supervised Regularized Multidimensional Scaling

Abstract

Iris Recognition is regarded as the most reliable and accurate biometric identification system available. In Iris Recognition, a person is identified by the iris region of the eye using image processing, pattern matching and the concept of neural networks. A typical Iris Recognition system involves three steps, Iris pre-processing, Iris feature extraction and Iris Classification. Most of the researchers use Daugman’s integro-differential operator and Daugman’s rubber sheet model for pre-processing. A number of feature extraction methods can be used to achieve a reasonable recognition rate. In our work we have used Supervised Regularized Multidimensional Scaling proposed recently for feature extraction that is used directly on iris image regarded as high dimensional vector. The method uses radial basis function to select some images as centres and then projects higher dimensional vectors into a lower dimensional space using an Iterative majorization algorithm. The projection is done in such a way that data of same class projects together and also it selects the most effective features that leads to better recognition rate. This approach excludes the pre-processing that saves computation time. We have compared our approach with Principal Component Analysis and implemented on a benchmarking data MMU iris data. K-Nearest Neighbor classifier is used for the classification. Numerical experiments show that Supervised Regularized Multidimensional Scaling successfully achieves better recognition and outperforms some other approaches such as Principal Component Analysis with and without pre-processing of iris images.