Hybrid Methodology for Blood Vessel Segmentation on Fundus Image: Frangi Filter, Otsu Thresholding and Morphological Operations

Abstract

Retinal vascular disease remains a significant medical concern, yet accurately segmenting blood vessels in fundus images continues to be challenging due to uneven lighting, low contrast, and the wide variability in vessel morphology—including very fine capillaries as well as broad arteries and veins .Automated segmentation not only improves diagnostic precision but also significantly reduces the manual workload of ophthalmologists, enhancing efficiency in both clinical and large-scale screening settings. This study aims to segment retinal blood vessels through a three-stage framework: Preprocessing: Improve image quality by applying CLAHE (Contrast Limited Adaptive Histogram Equalization) and a median filter to the green channel. Segmentation: Combine multiple techniques—Frangi filtering, 2D convolution, additional median filtering, Otsu’s thresholding, morphological operations, and background subtraction—to robustly delineate vessel structures. The proposed model is evaluated using statistical parameters on images from two publicly available databases. We achieve average accuracies of 0.9418 and 0.9086 for DRIVE and STARE databases, respectively. These metrics indicate that the proposed model is an effective and viable alternative for retinal vessel segmentation, offering a strong balance of precision and practicality for clinical and research applications.