Résumé en anglais

Geodesic methods have been widely applied to image analysis [17]. In this paper, we propose an automatic anisotropic fast marching based geodesic method to extract the centrelines of retinal vessel segments and their boundaries. Our method is related to the geodesic or minimal path technique which is particularly efficient to extract a tubular shape, such as a blood vessel. The proposed method consists of a set of pairs of points. Each pair of points provides the Initial point and Target point for one geodesic. For each pair of Initial point and Target point, we calculate a special Riemannian metric with an additional Radius dimension to constrain the fast marching propagation so that our method can get a nice path without any shortcut. The given pairs of points can be easily obtained from a pre-segmented skeletonized image by any vessel detection filter like Hessian or Oriented Flux method. Experimental results demonstrate that our method can extract vessel segments at a finer scale, with increased accuracy.