Retinal images are frequently corrupted by unwanted variations in intensity that occur due to general imperfections in the image acquisition process. This inhomogeneous illumination across the retina can limit the useful information accessible within the acquired image. Specifically, this can lead to serious difficulties when performing image processing tasks requiring quantitative analysis of features present on the retina. Given that the spatial frequency content of the shading profile often overlaps with that of retinal features, retrospectively correcting for inhomogeneous illumination while maintaining the radiometric fidelity of the real data can be challenging. This paper describes a simple method for obtaining an estimate of the illumination profile in retinal images, with the particular goal of minimizing its influence upon features of interest. This is achieved by making use of Laplace interpolation and a multiplicative image formation model.
展开▼
