A Novel Vessel Segmentation in Fundus Images Based on SVM

This paper proposes a novel retinal vessel segmentation based on supervised method in color fundus images. To determine whether a pixel is the vessel pixel or not, 30 features are extracted for each pixel in the fundus image which consisting of features in Gaussian scale space, features by multi-scale Gabor filter and divergence of vector fields. Then we use the features obtained by the previous process to train the Support Vector Machine (SVM) Classifier. Finally, the image pixels are classified as vessels and non-vessels using a non-linear SVM Classifier. We report experimental results on a public retinal database DRIVE, demonstrated an outperformance in comparison with other supervised retinal vessel segmentation methods. It is suitable for computer aided diagnosis.