Retinal vessel centerline extraction using multiscale matched filter and sparse representation-based classifier

Retina located in the back of the eye contains useful information in the diagnosis of certain diseases. By locating a blood vessel's width, color, reflectivity, tortuosity and abnormal branching, one can deduce the existence of these diseases. In order for this to be achieved, blood vessels first need to be extracted from its background in fundus image. In this paper we propose a new method to extract vessels based on Multiscale Production of Matched Filter (MPMF) and Sparse Representation-based Classifier (SRC). First, we locate vessel centerline candidates using multi-scale Gaussian filtering, scale production, double thresholding and centerline detection. Then, the candidates which are centerline pixels are classified with SRC. Particularly, two dictionary elements of vessel and non-vessel are used in the SRC process. Experimental results on two public databases show that the proposed method is good at distinguishing vessel from non-vessel objects and extracting the centerlines of small vessels.