Spectral Estimation of Printed Colors Using a Scanner, Conventional ‎Color Filters and Applying Backpropagation Neural Network ‎

R econstructing the spectral data of ‎color samples using conventional ‎color devices such as a digital ‎camera or scanner is always of ‎interest. Nowadays, multispectral ‎imaging has introduced a feasible ‎method to estimate the spectral ‎reflectance of the images utilizing ‎more than three-channel imaging. ‎The goal of this study is to ‎spectrally characterize a color ‎scanner using a set of conventional ‎color filters. To this end, a 1355 ‎chart was generated and printed; ‎the images of the printed charts ‎were scanned putting a translucent ‎color filter in front of each page ‎during scanning process. Each ‎page was scanned with 4 color ‎filters, including gray, blue, green ‎and yellow ones. A feed-forward ‎Back-Propagation neural network ‎with 12 input neurons of camera ‎responses, one hidden layer ‎containing 20 neurons, and an ‎output layer of 31 neurons of ‎spectral reflectance values was ‎applied. It was shown that it is ‎accurately possible to estimate the ‎spectral data of printed samples ‎from the scanner responses using ‎conventional color filters and the ‎proposed NN with an average ‎GFC value of 0.999. The mean of ‎color difference error was about ‎‎0.612 CIEDE2000 (1:1:1) unit or ‎‎0.987 CIELAB unit.‎