Local entropy thresholding based fast retinal vessels segmentation by modifying matched filter

The retinal blood vessels are highly responsible for the detection of retinal pathology such as glucoma, hypertension, arteriosclerosis and diabetes. So the segmentation of retinal blood vessels from their background is a prominent task. The objective of this paper is to present an automatic local entropy thresholding based fast, efficient and accurate retinal blood vessels segmentation method by modifying the standard Gaussian shaped matched filter reported in other papers in literature. Another objective is to identify the thin blood vessels together with large blood vessel segments, which is not considering in some existing blood vessels segmentation methods in literature. The proposed method has been implemented on forty retinal images taken from DRIVE database and segmented results are compared with hand-labeled ground truth images also available in the DRIVE database. The efficacy of the proposed method was examined and presented in terms of overall sensitivity, specificity and accuracy. Further, the performance of the proposed algorithm is compared with some other existing standard methods for the same task available in literature and the performance of the proposed method is found to be performing significantly better.