Study of retinal biometric systems with respect to feature classification for recognition and diabetic retinopathy

A biometric system is a heterogeneous entity characterized by identification and classification of data depending on a metric along with characteristics such as disease recognition and classification. This work is mainly concentrated on biometric pattern matching of a retinal image based on its macro and micro features. During data collection, several instances of abnormalities were detected on the retina, specially diabetes. In this work 350 images (fundus data which are training data) were analyzed using Decision trees and K Nearest Neighbour clustering technique. Four groups were identified from the data: normal retinopathy, mild non-proliferative diabetic retinopathy, moderate non-proliferative diabetic retinopathy and severe non-proliferative diabetic retinopathy. The important features that were collected for the data were distances, micro-aneurysms, exudates and haemorrhages. These features were extracted depending on the shapes of the vessels and the shapes of the diabetic features. In this paper a constructive method has been designed for classifying a person based on retinal biometric.The percentage accuracy by using decision tree has been 91.3% and by using K Nearest Neighbour has been 62%.The size of the training data set has been 350 and the size of the testing data set is 150.