Unsupervised Vessel Segmentation Method in Retinal Images

Abstract

This study presents a low complexity and an automated retinal segmentation approach. This technique processes the G-channel. Later, CLAHE, the PCA, and the Matched Filters are applied. Finally, segmentation is achieved using Otsu’s thresholding. Our technique is assessed on DRIVE and STARE databases. Simulations show that our method obtains accuracy of 95.26% on DRIVE and 94.55% on STARE. Our technique consumes less than 1 second on conventional machine to yield the segmented output image.