Iris recognition using support vector machines

In this thesis, an iris recognition system is presented as a biometrically based technology for person identification using support vector machines (SVM). We propose two approaches for iris recognition, namely: The approach I, which is based on the whole information of iris region and the approach II, where only the zigzag collarette region is used for recognition. In approach I, Canny edge detection and Hough transform are used to find the iris/pupil boundary from eye's digital image. The rubber sheet model is applied to normalize the segmented iris image, Gabor wavelet technique is deployed to extract the deterministic features and the traditional SVM is used for iris patterns classification. In approach II, an iris recognition method is proposed using a novel iris segmentation scheme based on chain code and zigzag collarette area. The Multi-Objectives Genetic Algorithm (MOGA) is employed to select features extracted from the normalized collarette region by log-Gabor filters to increase the overall recognition accuracy. The traditional SVM is modified to asymmetrical SVM to treat False Accept and False Reject differently. Our experimental results indicate that the performance of SVM as a classifier is better than the performance of classifiers based on feed-forward neural network using backpropagation and Levenberg-Marquardt rule, K-nearest neighbor, and Hamming distance