Iris Recognition System Using Principal Component Analysis based Back-Propagation Learning Neural Networks

Abstract

This paper deals with the iris recognition system using Back-Propagation learning neural network algorithm where Principal Component Analysis technique has been used to reduce the dimension of the iris feature vector. Automated iris localization and segmentation methods have applied to effectively isolate the iris region from pupil and sclera. Circular Hogue transform has used to detect the iris/sclera boundary and pupil/iris boundary. To remove the upper and lower eyelids effects from the iris, linear Hogue transform has used. Daugman's Rubber Sheet Model has been applied to normalize the iris pattern into a fixed dimension. To extract the features from the iris pattern, Log-Gabor filtering technique has used which extracts 9600 feature values for each iris pattern. Principal Component Analysis based dimensionality reduction technique has been used to reduce the feature into 550 dimensions. Finally, reducedfeatures are fed into the Back-Propagation learning neural network for learning. The recognition of unknown iris pattern is then performed by comparing this special pattern to the pattern for which an iris learned module is already built. CASIA-IrisV4 database has used to measure the accuracy of the proposed system. Experimental results and performance analysis show the versatility of the proposed iris recognition system.

Keywords: Iris recognition, iris segmentation, log Gabor filter, principal component analysis, Back-Propagation neural network