A Hybrid Approach for Automatic Exudates Detection in Eye Fundus Image

Diabetic retinopathy is an eye disease that is caused by damaging of the retinal blood vessels due to long term diabetes. It is important to automatically detect the DR lesions at early stage in order to prevent the further vision loss. Exudates are bright lesions that are considered as primary sign of this disease. In this paper, a hybrid approach is proposed for automatic detection of exudates from eye fundus images. In this approach, unsharp masking is used for preprocessing, region based segmentation is used for candidate detection and then pixel based classification is used to determine the severity level of the disease. The proposed method is tested the image-level as well as at pixellevel on publically available database DIARETDB1 that contain total 89 fundus images and results in 98.12 % specificity and 90.83% sensitivity considering the pixel-level evaluation, while it is 86.04% and 91.06% respectively on image-level evaluation .