Segmentation of retinal blood vessels based on analysis of the hessian matrix and Clustering Algorithm

In this paper, a novel unsupervised method to segment retinal blood vessels from colour fundus images is proposed. A new vesselness measure is introduced which is based on detecting vessel centerlines and orientation in scale space. Based on this vesselness measure a generated ground truth (GGT) image is obtained by thresholding and removing segments of small sizes. The segmentation is obtained by using this GGT image in conjunction with a RAdius-based Clustering Algorithm (RACAL). A dataset of 20 images publicly available is used to evaluate the performance of our proposed method. Experimental results show that a true positive rate (TPR) of 81% at false positive rate (FPR) of 4.5% is achieved compared with TPR of 76% at the same FPR from the piecewise threshold probing method [1].