Neural network based iris pattern recognition system using discrete Walsh Hadamard transform features

An iris pattern recognition system is developed for CASIA database using discrete 2D Walsh Hadamard (WH) transform as a tool for feature extraction. Experimental prototype pattern recognition (PR) system is designed in which iris images of ten different persons are given as input to the system. After localizing region of interest (ROI), features are extracted with respect to image statistics, texture and 2D WH transform domain. Based upon these features, an optimal feature vector comprising of only 23 features is selected and it is given as input to neural network based Pattern Recognition (PR) system. Two different neural network configurations including Multi Layer Perceptron (MLP), Radial Basis Function (RBF) and a different learning machine, known as Support Vector Machine (SVM) are investigated for their suitability as a PR system. It is observed that MLP neural network based PR system comprising of only one hidden layer containing 20 processing elements (PEs) outperforms others in respect of performance measures on cross-validation (CV) dataset.