A Study on the Iris Biometric Authentication

This paper presents a methodology for quantitatively establishing the discriminative power of iris biometric data. It is difficult, however, to establish that any biometric modality is capable of distinguishing every person because the classification task has an extremely large and unspecified number of classes. The purpose of this study is to investigate various combinations of features, distance measures, and classifiers to find the best combination for determining the individuality of the iris biometric. Based on this review, a proposed methodology is used to establish a measure of discrimination that is statistically inferable. To establish the inherent distinctness of the classes, i.e., to validate individuality, we transform the many class problem into a dichotomy by using a distance measure between two samples of the same class and between those of two different classes. Various features, distance measures, and classifiers are reviewed and evaluated. For feature extraction I compare simple binary and multi-level dimensional wavelet features. For distance measures I examine scalar distances, feature vector distances, and histogram distances. Especially, I compare conventional binary feature distance measures and evaluate their performance. Finally, for the classifiers I compare Bayes decision rule, nearest neighbor, artificial neural network, and support vector machines. The experiment of the eleven different combinations is performed. The best one uses multi-level 2D wavelet features, the histogram distance, and a support vector machine classifier.