Robust Iris Classification through a Combination of Kernel Discriminant Analysis and Parzen Based Probabilistic Neural Networks

Iris template classification in unconstrained environment is one of the open challenges in recognizing human through iris biometric modality. The iris template classifier must be robust to the outliers and noise introduced in the individual iris class distribution because of the occlusion, blur, specular reflection, etc. Also, it should perform fast enough, to make its use in real-world applications. We are introducing a combination of feature reduction technique called kernel discriminant analysis and parzen-based probabilistic neural network classifier which shows robustness to the outliers and noises and gives great advantage in time complexity as compare to other state-of-the-art classifiers. Comparisons are presented with the state-of-the-art classifiers like Euclidean distance, hamming distance with mask template, support vector machine, and sparse representation based classifier on two publicly available iris databases: CASIA-Iris-Thousand and CASIA-Iris-Lamp.