Retinal vessel segmentation for medical diagnosis

Abstract

This paper proposes the color fusion method, a supervised method for segmenting the retinal vessels. This method uses the feature fusion with dimensionality reduction, FFdr. The feature vectors are extracted from RGB channels using five feature extraction methods. The classification is done by a support vector machine (SVM) applying both linear and non-linear functions. The DRIVE database which holds colored retinal images together with precisely segmented vessel images by experts is used to evaluate the proposed method. Comparing to the existing methods in literature, it performs the second best in terms of accuracy and sensitivity with the best average accuracy of 0.9506. It has the desirable minimum false positive rate. Its effectiveness and performance are demonstrated via receiver operating characteristic analysis.