Iris recognition system for some clinical applications

Abstract

Today, with the increase in security threats all over the world authentication of an individual is becoming an important issue and area of interest for researchers. Over the traditional password or key based security systems biometric authentication systems are considered as very accurate and reliable. Iris Recognition System is one of them. It is very accurate system as iris images of twins or iris images of even left and right eye of same person are different. Numerous researchers have given iris recognition systems based on different feature extraction techniques. In this work, a comparative study of the existing techniques has been carried out. A simple, fast and effective statistical feature extraction based iris recognition system has been proposed and implemented. Features have been extracted in two different directions, namely, radial direction and angular direction. An attempt has been made to study the effect of number of features as well as the radial and angular resolution while normalization. Results obtained are effective, encouraging and comparable to existing techniques. In literature, little work has been reported on clinical applications of iris recognition systems. In this thesis, clinical applications of iris recognition system have also been investigated. Three different applications, i.e., to predict the gender of imposters, to predict diabetes and to predict obstructive lung disease have been considered. In security systems predicting gender of an imposter is equally important to determine the identity. Most of the work to predict the gender utilized facial images. A few studies have been reported using iris images. In the present research work, Support Vector Machine (SVM) based gender prediction model has been proposed and implemented. Results obtained show the effectiveness of system over the existing models. Further, a non-invasive and non-contact type model, i.e., a system as an aid to doctors is proposed to predict the disease from iris images. Iridology is a science to determine the status of health of an individual from iris patterns. It does not diagnose the disease. If it is known that which part of human body is not well then by doing regular exercises and following healthy habits one can delay if not prevent the occurrence of disease. iv In this thesis, iris recognition algorithm has been combined with iridology to predict diabetes and obstructive lung disease. A dataset of healthy subjects and also the subjects suffering from diabetes has been created for implementing the problem of diabetes prediction. Another dataset of healthy subjects and also the subjects with obstructive lung disease has been created. These datasets have been built with the help of I-SCAN-2 dual iris scanner of Cross Match Technologies Inc. In order to develop the models for the two clinical applications, features have been extracted using two different techniques: wavelet transform and Gabor filter. SVM based classifier has been implemented to predict the subject as healthy or suffering from diabetes or obstructive lung disease. The maximum overall accuracy of 88.3% and 89.0% has been achieved for predicting diabetes and obstructive lung disease, respectively. This accuracy has been achieved for a test data of size 80 and 100, respectively for the two applications. The proposed models are compact and handy.