Retinal blood vessel classification based on color and directional features in fundus images

The symptoms of some diseases such as high blood pressure and diabetic retinopathy affect on the retinal vessels can be helpful to control the progress of these diseases. In this paper, our aim is to detect and classify the retinal vessels to arteries and veins. This algorithm achieves the vascular tree structure using a local entropy-based thresholding segmentation method. Next, several color and novel directional structural features are extracted. The structural features are based on wavelet, projection and profile of vessels. Then, Principal Components Analysis (PCA) algorithm is used for optimizing the extracted features. Finally, the vessels are classified by a neural network classifier. By using the results of our optimization algorithm in the feature selection, we achieved high sensitivity and specificity and generally, the accuracy rate of 92.9% was obtained on the test dataset.