Retinal vessel profiling based on four piecewise Gaussian model

Retinal vessel profiling in ocular fundus images is a basic tool to measure and detect retinal vascular structure, which can provide important information to clinical diagnosis and medical research. In this paper, a four-piecewise Gaussian model is proposed to describe the intensity distribution of retinal vascular cross-section. Eight hundred seventy three vessel profiles from 34 color fundus images were tested and best-fit parameters were obtained using Levenberg-Marquardt method. The mean squared error of the four-piecewise Gaussian model is about 0.18. The proposed model can describe the vessel profile with or without central reflex, and the profile with non-uniform background well.