Extraction of vascular system in retina images using Averaged One-Dependence Estimators and orientation estimation in Hidden Markov Random Fields

The proper segmentation of the vascular system of the retina currently attracts wide interest. As a precious outcome, a successful segmentation may lead to the improvement of automatic screening systems. Namely, the detection of the vessels helps the localization of other anatomical parts and lesions besides the vascular disorders. In this paper, we recommend a novel approach for the segmentation of the vascular system in retina images, based on Hidden Markov Random Fields (HMRF). We extend the optimization problem of HMRF models considering the tangent vector field of the image to enhance the connectivity of the vascular system consisting of elongated structures. To enhance the probability estimation during the solution of the Hidden Markov problem, the Averaged One-Dependence Estimator (AODE) is used instead of the commonly used naive Bayes estimators, since AODE uses a weaker assumption than total independence of features. The advantages of our method is discussed through a quantitative analysis on a publicly available database.