Spline-based refinement of contours in binary maps of retinal vessels

Vasculature segmentation and vessel caliber measurements in retinal images can improve early diagnosis of several diseases, such as diabetes, retinopathy of prematurity and hypertension. The aim of this thesis is to present a novel algorithm for improving the vessel contours obtained from binary vessel maps. This is useful for quantitative evaluations like width and tortuosity estimation. Two algorithms are described in this document. Firstly, a simple vessel segmentation strategy filtering the image using a Gaussian kernel and producing a binary vessel mask from the response image by the application of a thresholding step. Secondly, a procedure fitting the two contours of each vessel in the binary map with a cubic spline curve, under a parallelism constraint between the two splines. The second algorithm is the main focus of this work. The performance of the algorithm has been evaluated on the publicly available REVIEW database, which contains a set of images with vasculature showing different characteristics. Images also include several manual measurements made by three independent observers. The method is implemented in MATLAB