Diabetic Retinopathy (DR) is the most common disease induced by the complication of diabetes,
causing blindness. In many rural areas, the contributions of ophthalmologists are predicatively
less to treat the disease. Detection of lesions in the early stage is a progressive measure to diagnose
DR. Initially, a preprocessing method is performed to detect the Optic Nerve Head (ONH) in the lesion.
Based on the degree of reflectance in ONH, feature extraction is computed using multi-scale Local
Binary Pattern (LBP) algorithm. Here, Gabor convolution is estimated and the structure of ONH is
encoded. This extends to a statistical computation in terms of the moment and standard deviation. A Support Vector Machine
(SVM) classification is formulated to locate the hemorrhages and exudates and an effective probabilistic multi-label
lesion classification is performed to acquire five sets of results representing the diabetic retinopathy: 1) Grade-1 Exudates,
2) Grade-2 Exudates, 3) Micro aneurysms, 4) Hemorrhages, 5) Neovascularization. Finally, the affected area of lesions is
used to diagnose the disease.
Keywords: Gabor convolution, Local Binary Pattern, Optic Nerve Head, Statistical computation, Support Vector Machine.
Rights & PermissionsPrintExport