Özet:

Abstract. Diabetic retinopathy is a chronic retinal disorder that affects nearly all patients with diabetes mellitus, albeit with different severity. In this paper a new method based on graph matching is offered. Mentioned method has three main phases which in the first phase, entered image is preprocessed in order to become an image with uniform background and appropriate contrast. Then image is segmented by matched filter with Gaussian-Hermite kernel and retinal vessels will be extracted. In the second phase vascular structure graph is created with fractionating and in the last phase neural network is used for diagnosing normal images and images with signs of diabetic retinopathy. Images that are used for recommended method are derived from the ImageRet database. This offered method has appropriate accuracy for distinguishing healthy and unhealthy images.

Anahtar Kelimeler: