Feature extraction in retinal fundus images

An automated method for Retinal feature extraction is done for the diagnosis of several retinal diseases such as Diabetic Retinopathy (DR). In this paper, a new adaptive histogram approach is discussed to extract the features. Detection of Optic disk (OD) forms an important step in the automatic screening process for risk population at very low cost. OD extraction is quite difficult, since its brightness, color and contrast of OD is similar to other components such as cotton wools & exudates. OD serves as the landmark for fundus feature, namely fovea, macula and blood vessels, have predefined constant distance from OD center. The proposed method includes binary thresholding of fundus images followed by morphological operation to remove small objects. Then, the boundary of OD is traced using boundary tracing technique. The method is tested in the fundus images, from the screening section of Mahatma Gandhi Eye hospital & Research center, Tamilnadu. The centroid, perimeter and area for detected OD is evaluated &tabulated for further diagnosis. The evaluated value is appreciable and comparable to the standard methods available in literature.