Impact of Different Degree of Smoothing on Non-Local Means based Filter for Retinal Vessel Modeling

One of the complications of diabetes is the disruption of vision because of the damage to the retinal blood vessels, which is known as diabetic retinopathy (DR). The change in diameter of the retinal blood vessel is an important sign of DR. Several segmentation models have been proposed for measuring the retinal blood vessel diameter, but the results are not always satisfactory. Preprocessing steps, such as image filtering for noise reduction, can improve the result of the modeling step. This paper presents the results of applying non-local means based filter for retinal image enhancement to improve the performance of retinal vessel modeling using two-dimensional (2D) Gaussian. Various degrees of smoothing (DoS) of non-local means based filter were compared and applied to the retinal images from High-Resolution Fundus (HRF) public dataset. The results show that DoS value, which is 12, gave the highest R
²
(coefficient of determination) value, which is 0,9862. The proposed filtering approach yields encouraging results for retinal vessel modeling improvement.