Unsupervised Segmentation of Retinal Blood Vessels Using a Single Parameter Vesselness Measure

In this paper, a novel vesselness measure based on analysis of the Hessian matrix is presented. The larger eigenvalue of the Hessian matrix is used for vessel centerlines detection, while vessel orientations are estimated from the eigenvectors corresponding to the smaller eigenvalue. The vesselness measure combines information from vessel centerlines and orientations over scales to segment retinal blood vessels from colour fundus images. A publicly available dataset is used to evaluate the performance of our proposed method which has the advantage of being unsupervised and of using only one parameter.