Diabetic retinopathy is the one of the leading causes of vision loss across the globe, within increase in large number of diabetes mellitus patients. Early diagnosis and treatment are very essential for fast recovery. Non-proliferative diabetic retinopathy is the early stage of disease, identified by leaky and irregular shaped blood vessels. Colour fundus imaging technology is widely used by ophthalmologists to determine the nature of disease. The presence of noise and non-linear illumination are the major challenges. Image processing techniques could enhance the image and extract the various features of disease like the presence of microaneurysms, hemorrhages, hard exudates and soft exudates, followed by quantification of these features. In this paper, a fuzzy logic-based classifier is designed for the classification of non-proliferative diabetic retinopathy as normal, mild, moderate, severe and advanced from colour fundus images. The input to fuzzy inference system (mamdani type) are the extracted features from fundus image. Fuzzy rules judge the severity of disease by considering the true pixels from the input feature maps. This design is implemented using MATLAB code and results shows a high degree of accuracy in classification.

Diabetic retinopathy, colour fundus imaging, Classification, Fuzzy Logic.