Influence of the Size of the Training Set on Colour Measurements Performed using a Multispectral Imaging System

摘要

The accuracy of multispectral imaging systems for recovering reflectance spectra and/or measuring colour has been widely studied and a great deal of research work has been carried out to find the best set of filters and characterization methods. Less work has been carried out on the training set of colour patches used to characterize the system.rnThe main aim of this work is analyzing the influence of the size of the training set on colour measurements performed using an imaging system based on a CCD camera. In order to perform this study, training sets of different size were obtained using an easy selection method, based on differences in a~* and b~* CIELAB coordinates between the selected colour samples from an initial training set.rnTwo configuration of the imaging system have been used: a colorimetric configuration with 3 acquisition channels, and a multispectral configuration with 7 acquisition channels. Very similar results were obtained for both configurations, and it has been proved that the improvement in accuracy of colour measurements by increasing the size of the training set is limited, probably due to the fact that the expected improvement due to the increase of information is made up by the increase of errors accumulated in adjustment. Furthermore, a proper selection, uniform over the colour gamut available, of a low number of samples (40) allows to characterize the imaging system as well as a greater set of samples, for instance, the whole useful set of colour patches (166) of the GretagMacbeth ColorChecker DC chart.