Illumination correction and contrast equalization in colour fundus images

Abstract

colour fundus images based on the B-spline approximation of the illumination surface is presented. The control points for B-splines are determined from an original image, sepa- rately from the red, green and blue channel. The estimated illumination surface is used in a multiplicative model for fast correction. The proposed approach has been subjec- tively tested on new publicly available retinal image database DRIIL (Digital Retinal Images for ILumination correction), which contains 165 images. Objective testing has been con- ducted by investigating the influence of the proposed method on blood-vessel segmentation using another high resolution fundus image dataset of healthy and diabetic patients.