A Novel Fuzzy-Based Thresholding Approach for Blood Vessel Segmentation from Fundus Image

Abstract

Retinal vessel segmentation is a vital part of pathological analysis in Fundus imaging. The automatic detection of blood vessels resolves several issues in the manual segmentation process. Most unsupervised segmentation methods depend on conventional thresholding techniques for final vessel extraction. It may lead to the loss of some vessel pixels, leading to inaccurate analysis of retinal diseases. In this work, we incorporate fuzzy concepts into two threshold-based vessel detection methods, namely “mean-c thresholding” and “Iso-Data thresholding,” which results in a mask consisting of membership values rather than binary values. The two fuzzy-based thresholding algorithms are applied independently on each image, and the resultant membership image (mask) is fused to get a single membership mask. The fusion is performed using fuzzy union operation. Experiments are carried out with Fundus images from DRIVE, STARE and CHASE_DB1 databases.ses. The proposed fusion framework gives a 3%, 6%, and 5% increase in sensitivity compared to traditional thresholding methods when applied to the DRIVE, STARE, and CHASE_DB1 databases, respectively. The accuracy obtained for the datasets is 96.02%, 94.57%, and 94.34%, respectively.