Detection of diabetic retinopathy in fundus images using vector quantization technique

Diabetic retinopathy is one of the major cause of blindness among the people. Many approaches are proposed by the authors to automate and detect the presence of diabetic retinopathy in fundus image. We propose a novel method of detection of the diabetic retinopathy using Gaussian Intensity feature input to a VQ classifier. The underlying idea of using this technique in fundus imaging is that there are certain features which pertain only to diabetic retinopathy. These features are extracted in terms of diameter of the blood vessels expressed by sigma, and the height of the gaussian profile across the cross-section, given by h. In this work, 30 images are taken as normal images and 25 pathological images are considered. A successful average diagnostic performance of 90% is achieved in this method.