Retinal Blood Vessel Segmentation Based on Vessel Branch Width Adaptation

This paper presents a new blood vessel segmentation approach in retinal images. Initially, an enhancement method based on illumination and contrast adjustment along with Gaussian smoothing is used in preprocessing step. Local coarse vessel segmentation and vessel refinement are employed in the next step to retrieve the vessels of various thicknesses. Circle Test and Branching Segment Detection (BSD) methods are reintroduced in this paper to adjust the algorithm with the width of vessel branches. The proposed method is evaluated with a recently popular measure BAcc (Balanced Accuracy), on DRIVE database. The results show that the proposed method outperforms most of unsupervised retinal blood vessel segmentation algorithms.