Automatic Extraction of Micro-aneurysms and Haemorrhages from Digital Fundus Image

In Diabetic Retinopathy, red lesions are consisting of micro-aneurysms and haemorrhages. The paper deals with the proper detection of micro-aneurysms and haemorrhages which are found in fundus images using an automated computer vision. Morphological operations are performed to extract out all the possible candidates that have similar pixel intensity as that of the red lesions. To reject the blood vessels effectively, Gabor filter is used in this paper. Discriminatory features are extracted and fed to train SVM classifier for the proper classification of the micro-aneurysms and haemorrhages. The algorithm developed is tested on 168 fundus images taken from DIARETDBI AND MESSIDOR databases. It achieved an overall accuracy of 93% and 91.8% in classifying the micro-aneurysms and haemorrhage respectively. Proposed work is efficient and the result are encouraging to use in real time applications.