Performance Analysis of Gray Scale and Color Iris with Multidomain Feature Normalization and Dimensionality Reduction
Iris recognition techniques have come a long way since the preliminary proposal by Daugman. There are several techniques which mainly focus on improving the performance of IRIS recognition system either with the help of improved classifier or the feature set. However we have proved that the performance of IRIS recognition to a large extend depends upon the inter dependability and separability of the feature vectors from different classes in the feature plane. If such dependency is identified and dimensions or the features for which most of the classes represent similar and inseparable values. This finding lead us to investigate the dimension reductionality on the feature vectors. Most of the Iris recognition technique still relies on Gabor filter. But modern IRIS recognition sensors present a great deal of details about the IRIS part and the resultant images are color image. Hence there is a need to analyze the behavior of dimensionality reduction techniques for both gray scale conventional IRIS recognition technique as well as the one applied on color images. In this work we apply the dimensionality reduction technique on multi-domain features extracted from the images for set of color and gray scale IRIS dataset. The result shows that the dimensionality reduction on color images improves the performance of the classifier by .8% and the performance of the gray scale image database classifier is improved by .3%.
