Segmentation of Retinal Blood Vessels based on Adaptive Interval Top-Hat Filtering in Heterogeneous Environment Optic Disc Image

摘要

Blood vessel segmentation on a retinal image is one of the algorithms for identifying retinal diseases and one of the essential steps for various ocular imaging applications. This paper proposes methods for segmenting vessels on the optic disc area of a retinal image. The method consists of 3 main phases. First, vessel enhancement is performed on the green channel extracted from RGB fundus image. Second, the cartesian image is transformed to polar space for improvement of the segmentation process. Third, the vessels are obtained using a Multi-stage Adaptive Top-hat transformation. The algorithm is evaluated using confusion metrics, sensitivity, specificity, precision and accuracy. The performance reaches 96.11% of accuracy, 80.31% of precision, 66.12% of sensitivity and 97.89% of specificity.