Segmentation of retinal blood vessels by using a matched filter based on second derivative of Gaussian

Abstract

An accurate retinal blood vessel segmentation is a prominent task in computer aided diagnosis of various retinal pathology such as hypertension, diabetes, glaucoma, etc. The matched filter based retinal blood vessel segmentation approaches are simple yet effective. However, matched filter based approach detect both vessels and non-vessel edges. Hence, this also leads to false vessels i.e. non-vessels detection. To overcome the problem of detecting the non-vessel edges, here we propose an extension of matched filter based on the second derivative of Gaussian (SDOG-MF). The proposed approach is simple and effective for the segmentation of thin as well as thick retinal blood vessels. The experimental results obtained for both DRIVE and STARE databases confirms that the proposed method has higher TPR, FPR, and accuracy as compared with other available retinal blood vessel segmentation approaches in literature. Further, the performance of the proposed method is also better for the segmentation of pathological retinal images.