Effects of uneven channel image illumination on retinal image spectra

Abstract

Spectral retinal imaging is a potential technology to renew the diagnosis of diabetic retinopathy and other eye-related diseases causing abnormalities in the retina. Based on the retinal images, the changes can be diagnosed in an automated manner using image analysis. This thesis presents a review of the state-of-the-art methods for correcting of spectral retinal image illumination fields and studies the existing effects of uneven illumination fields on retinal image spectra. The study on the effects is based on the information extracted from the retinal blood vessels, which are segmented and traced from the retinal images with a convolutional neural network and with ARIA tracing. Experiments were made with a dataset containing spectral retinal images with different disease-related abnormalities. The experiments show the dependence between the uneven illumination fields and changes in the retinal image spectra.