Detection of retinal blood vessels and reduction of false microaneurysms for diagnosis of diabetic retinopathy

Diabetic retinopathy affects the human eye and causes the blindness. For the efficient diagnosis of retinopathy accurate measurement (true vessel structure) of vessel diameter is required for estimating the types of vessels. True approximation of total number of microaneurysms is required for estimating the stages of diabetic retinopathy. This work presents an automated system for detection and prediction of diabetic retinopathy severity (based on stages) on retinal fundus image. The algorithm starts by preprocessing the image by spatial low pass filter and for feature extraction optimized Gabor filter is used. Further integrated approach of morphological operation erosion and extended minima transform is used for estimating true vessels structure. An approach of Skelotonization is also proposed for estimation of true vessel structure. Euclidian distance measure approach is applied for calculating the diameter of vessels at four discrete points in filtered image. Stages of diabetic retinopathy are classified based on calculated diameter.