An Improved Retinal Blood Vessel Segmentation Algorithm based on Multistructure Elements Morphology

Retina is the portion where many important eye diseases and systemic diseases manifest. By evaluating the retinal blood vessels, doctors can diagnose the primary stages of diabetic retinopathy, age related macular degeneration, glaucoma etc which may eventually lead to blindness. The objective is to develop an algorithm that segments the retinal blood vessels in a short time and with high accuracy. But the low gray level contrast and dynamic range of the image make the blood vessel segmentation process very difficult. A new multiscale transform, Curvelet transform, is used for retinal image contrast enhancement. Since the blood vessels are distributed in various directions multistructure elements morphology is used to find the blood vessel edges. The false edges are removed by morphological reconstruction. A locally applied level dependent thresholding algorithm with connected component analysis and length filtering removes the remaining false edges after reconstruction step. The proposed algorithm, when experimentally applied on images from the DRIVE database, gave an accuracy of more than 97% in less than 15 s, thus showing its effectiveness in retinal blood vessel segmentation.