The inverse problem of spectral reflection prediction: Problems of framework selection

Abstract

Digital image processing requires substantial computations for characterization. It is since the reliable color reproduction can be achieved by establishing the correspondence between the spectral reflectance of the printed surface and the amounts of deposited inks. The processing is implemented by using different mathematical models. Most of the color prediction models engage some mathematical techniques to predict spectral reflectance for a mixture of colorants that are characterized by absorption and scattering during the light propagation. However, few attempts were made to make a model for prediction the colorants values based on an observing spectrum. This work is devoted to application of artificial neural network approach for solving the inverse problem of spectral reflection prediction. This task has been considered unsolvable as it involves solving a system of the linear differential equations, in which the number of unknowns exceeds the number of equations. Our attempt is based on the assumption that the prediction of the initial colorants from spectral data is possible by analogy with the work of the color perception system in humans. The aim of our study is to offer an approach to the framework selection. The model is built in Matlab and shows satisfactory prediction accuracy. © 2020 American Institute of Physics Inc.. All rights reserved.