Automatic Detection of Exudates and Hemorrhage in Fundus images

Abstract

Diabetic Retinopathy disease might lead to total loss of diabetic patient’s sight. In this research, we aim to detect the lesions of the Diabetic Retinopathy including Hemorrhage, and exudates. To be sure that the fundus images are in the same conditions of brightness, we applied Grey World approach on brightness channel then reconstruct the RGB. CLAHE and Unsharp filter are applied on reconstructed green channel and resulted in corrected green channel Gcor. The contrast stretching function is applied on the original green channel to result in stretched green channel Gs. The two images Gcor and Gs, are processed using thresholding and morphology operations to extract and localize optic disk, exudates, blood vessels, micro-aneurysm, Hemorrhage and macula. The color correction steps highlights Diabetic Retinopathy lesions. Sensitivity in our computer Diabetic Retinopathy diagnoses system achieved 94.44 % and outperforms most of current research works.