A Novel Retinal Vessel Segmentation Method Using Connected Domain Merging and Improved Graph Cut

Retinal vessel segmentation is a key step in detecting diseases according to the retina, and the accuracy of blood vessel segmentation directly affects disease detection results. This paper proposes a novel algorithm to achieve fine segmentation of blood vessels, and the algorithm can also adapt to changes in the illumination of fundus images. Firstly, homomorphic filter is used to reduce the influence of non-uniform illumination. Secondly, the top-hat transform and Connected-Domain Merging are used to obtain a rough vessel segmentation and generate the candidate vessel. Finally, candidate vessel is expanded by morphological dilation to improve the contrast between target area and background area, and then improved graph cut algorithm is used to finely segment blood vessel. The proposed method is evaluated on STARE, DRIVE, and LES-AV, and the experiment result proves the effectiveness of the method.