Retinal Vessel Segmentation Through Denoising and Mathematical Morphology

Automated retinal blood vessel segmentation plays an important role in the diagnosis and treatment of various cardiovascular and ophthalmologic diseases. In this paper, an unsupervised algorithm based on denoising and mathematical morphology is proposed to extract blood vessels from color fundus images. Specifically, our method consists of the following steps: (i) green channel extraction; (ii) non-local means denoising; (iii) vessel vasculature enhancement by means of a sum of black top-hat transforms; and (iv) image thresholding for the final segmentation. This method stands out for its simplicity, robustness to parameters change and low computational complexity. Experimental results on the publicly available database DRIVE show our method to be effective in segmenting blood vessels, achieving an accuracy comparable to that of unsupervised state-of-the-art methodologies.