Automatic Retinal Vascular Network Detection using Multi-Thresholding Approach based on Otsu

Due to complexities like nonhomogeneous illumination noise, vessel width variation and very low contrast of the small-width vessels in relation to the retinal fundus background, the suitability of automatic global thresholding methods for the segmentation of retinal vessel network remains a subject of on-going research. This paper investigates the use of a multi-thresholding approach based on Otsu global threshold method for the segmentation of the retinal vessel network in fundus images. An experimental study conducted on DRIVE database shows that the vessel network segmentation result obtained from the investigated method detects large vessels and some thin vessels in the retinal fundus images. When compared to some previous multi-thresholding approaches in the literature, the proposed method achieved higher average accuracy value of 95.51%, average sensitivity value of 70.20%, average specificity value of 97.96%. The proposed multi-thresholding method also showed an higher average accuracy value when it is compared with some supervised learning techniques in the literature.