On the compensation of uneven illumination in retinal images for restoration by means of blind deconvolution

Abstract

Retinal eye fundus images are used for diagnostic purposes, but despite controlled conditions in acquisition they often suffer from uneven illumination and blur. In this work, we propose the use of multi-channel blind deconvolution for the restoration of blurred retinal images. The estimation of an adequate point-spread function (PSF) is highly dependent on the registration of at least two images from the same retina, which undergo illumination compensation. We use the bi-dimensional empirical mode decomposition (BEMD) approach to model the illumination distribution as a sum of non-stationary signals. The BEMD approach enables an artifact-free compensation of the illumination in order to estimate an adequate PSF and carry out the best restoration possible. Encouraging experimental results show significant enhancement in the retinal images with increased contrast and visibility of subtle details like small blood vessels.