Evaluation of Algorithms for Segmentation of Retinal Blood Vessels

Assessment of retinal blood vessels plays crucial role in detection of many eye diseases. Extraction of blood vessels using image processing algorithms does the task of image registration. Thus Retinal blood vessels extraction is very important task. In this paper, we propose a modified approach for segmentation of blood vessels using matched filtering using two different values of variance parameter. Evaluation of other novel approaches for segmentation of retinal blood vessels is carried out and compared with proposed system. In first stage of algorithm, preprocessing methods are employed for enhancement of image quality and removal of noise. In second stage, classical matched filtering, proposed modified matched filtering, second derivative of Gaussian function filtering, K-means clustering and Fuzzy C-means clustering methods are used for blood vessels enhancement. Local entropy thresholding and length filtering is used for final segmentation of blood vessels in third stage. An automated system thus developed using each of algorithms stated earlier is tested on set of 40 ocular fundus images from DRIVE database and evaluated for sensitivity, specificity and accuracy. Sensitivity, specificity and accuracy of proposed method are found to be highest and equal to 80.63%, 97.16% and 96.84% respectively.