Auto Segmentation of Retinal Blood Vessel Image using 2-D Gaussian Filter and Its First and Second Order Derivative

Abstract

we are presenting a novel unsupervised way of segmenting the vasculature in images of retina. This technique employs a FODOG and SODOG 2-D Gaussian filters with the modified local entropic thresholding algorithm for segmenting vasculature of retinal images. The algorithm is implemented on STARE and DRIVE datasets which are freely available. The SODOG method gives better performance on both databases in comparison with other existing techniques. On STARE and DRIVE database, it achieves average segmentation accuracy of 96.37% and 96.82% respectively. The time taken by the proposed algorithm for processing each retinal image is very less compared to other existing supervised methods. The simpleness and fast execution makes the proposed algorithm appropriate for automated analysis of retinal images thereby aids in diagnosing the diabetic retinopathy.