Reconstruction of retinal spectra from RGB data using a RBF network

In comparison with the standard three-channel colour images, spectral retinal images provide more detailed information about the structure of the retina. However, the availability of spectral retinal images for the research and development of image analysis methods is limited. In this paper, we propose two approaches to reconstruct spectral retinal images based on common RGB images. The approaches make use of fuzzy c-means clustering to perform quantization of the image data, and the radial basis function network to learn the mapping from the three-component color representation to the spectral space. The dissimilarities between the reconstructed spectral images and the original ones are evaluated on a retinal image set with spectral and RGB images, and by using a standard spectral quality metric. The experimental results show that the proposed approaches are able to reconstruct spectral retinal images with a relatively high accuracy.