Computer method for tracking the centerline curve of the human retinal blood vessel

In this paper, we propose a mathematical approach for tracking the centerline curve in retinal images. First, the undirected topology graph of the blood vessel is extracted from the given image; this is performed after the binarization of the image. Then, we use a skeletonization algorithm in order to obtain the human retinal vascular tree. Next, we determinate the pixels classification (endpoints, bifurcation points, and interior points) and branches curve. Finally, we use three methods of reconciliation of the blood vessels curve to get a smooth curve, particularly insensitive to deformations that may taint the subject, as well as the recognition of the natural structures of the human retinal vascular tree. The results obtained for the three types of reconstruction are compared between them and with the geometrical structure of the vascular tree. We note that the cubic spline method is better than the other two methods in terms of average Root-Mean-Square Error (RMS) value 0.12 pixels and the average Absolute value of the Aaximal Error (AME) 0.57 pixel. Their advantages and disadvantages are discussed in relation to other methods proposed in the literature.