Segmentation and detection of the retinal vascular network using fast filtering

Abstract

Changes in retinal blood vessels are a characteristic sign of many retinal diseases. Therefore, the automatic segmentation of vessels is an essential element for the diagnosis of different ocular diseases. In this paper, we present a novel algorithm for the detection and the segmentation of the vascular network of blood vessels in fundus images. Our algorithm employs two mean linear filters using the convolutional kernel, one directional along a line and the second on a square region, in combination with thresholding. The proposed approach’s performance was tested on the public datasets DRIVE and STARE. Based on the test results, the mean segmentation accuracy, sensitivity, specificity and time complexity of retinal images in DRIVE are 94.27%, 97.01%, 66.20% and 1.63 s and for the STARE database, they are 93.41%, 95.54%, 66.55% and 2.13 s respectively. The proposed algorithm is simple and very fast. It achieved satisfactory mean segmentation accuracy with very low time complexity