Retinal Vessel Segmentation Based on Multi-scale Line Detection and Morphological Transformation

Retina is an important part of human eyes, analyses of retinal vessels are preferable to the early prevention, detection and treatment of cardiovascular diseases. In this paper, we propose an effective unsupervised segmentation method. Firstly, the image is preprocessed to reduce interference of external factors; Secondly, the preprocessed image is enhanced to suppress the optic disc and background interference; Finally, based on the method of multi-scale line detection, the blood vessels images are obtained. We evaluate our algorithm on the public datasets DRIVE and STARE, the experimental results show that the accuracy is about 95.90% and 96.11%, the sensitivity is 76.70% and 75.86%, the specificity is 95.20% and 95.56% respectively. Compared with other unsupervised methods, the proposed algorithm has a better segmentation effect.