AN IRIS RECOGNITION SYSTEM BASED ON MORPHOLOGICAL OPERATIONS AND BIT PLANE SLI

Abstract

An efficient and reliable recognition of an individual is one of the primary objectives of today’s high security oriented environment. Biometrics is a technique of measuring an individual’s characteristics like face, fingerprints, gait and iris, for its identity. The recognition is done based on some unique characteristic which cannot be transferred or lost. Iris is located in the human eye which consists of highly unique, rich and complex patterns. Iris recognition has proved to be the most accurate methods among the existing biometric technologies which paved its way to the challenging research in the field of digital image processing. The stages of an iris recognition system cover image acquisition, image pre-processing, feature extraction and matching of the iris texture. The performance of the system mainly depends on the removal of noise, correct localization of iris and feature extraction. In the thesis, pre-processing of the iris region in the image is accomplished by use of the morphological operators and bit plane slicing. The morphological techniques are based on a set theory and are known for their speed as well as accuracy in the image segmentation. The image is probed with a suitable structuring element to perform the operation of dilation and erosion. The iris outer boundary has been detected with the help of a circular summation of the pixel intensities. The experiments have been done on CASIA version 1 iris image dataset. The algorithm has achieved accurate pupil and efficient iris localization when implemented on the dataset. Bit plane slicing has been used for feature extraction after the normalization of the iris from Cartesian to polar coordinates. Normalization of the iris effective region has been achieved by taking the pupil centre as a reference point. The normalized image is then decomposed into its bit planes. The 8 bits of a 256 level, grayscale image can be decomposed into eight bit planes, i.e. bit plane ‘1’ to bit plane ‘8’. The higher order bits contribute significantly as compared to the lower order bits. Since the normalized image contains mainly the middle frequencies of the image due to the normalization process of iris so bit plane ‘5’ to bit plane ‘7’ are considered to contain the maximum information for the feature extraction. Bit planes 5 and 6 have been found to provide recognition rate of 98% and 92% respectively as compared to other bit planes for CASIA version 1 dataset. vi

T