A New Vessel Enhancement Transform on Retinal Blood Vessels Segmentation

Abstract

The automatic detection of retinal blood vessels is a very important task in computer aided-diagnosis of retinal diseases. In this work a new method is proposed for the automated detection of the retinal vessels. Three new and important contributions are made. First, a new method capable of vessel enhancement is presented. Second, a new criterium to remove some false vessels caused by the proximity to bright regions is presented, avoiding the false vessels created by the presence of exudates or bright artifacts. Third, a new method that discards the false vessel regions that usually tends to appear in the border of the optic disc. This is achieved using the derivatives of the gradient magnitude local maxima over different scales. The performance evaluation is made on two publicly available databases, namely, STARE, and HRF with state-of-the-art results. Particularly, the described method reveals to be very reliable on retinal images with large pathological signs.