Optimization of total multispectral imaging systems: best spectral match versus least observer metamerism

Abstract

The paper deals with total multispectral imaging systems assembled from a multispectral camera or scanner for capturing color images or natural scenes and a multichannel color image display for image synthesis. The aim of the multispectral technology is to reproduce an approximation to the spectral color stimuli of an original image in order to reduce color mismatches between original and reproduction for every human observer and any illuminant considered. Two system aspects are discussed in more detail: the representation of spectral data by a set of multispectral values at the interface between camera and display and the method of controlling the channels of the multichannel display to reproduce colors at least errors for every human observer. An expansion of spectral stimuli functions into a series of basis functions with weighting coefficients called multispectral values is used for the data representation at the interface. The set of basis functions is either optimized with respect to best fitting the original color stimuli or to produce least observer metamerism. It is shown that observer matched basis functions lead to smaller color errors. Best results for the control of the multichannel display are achieved with an iterative process using stochastic pattern generation. It is shown that maximum color errors for a large test set of spectral color stimuli on one side and 24 different human observers on the other are below 1,6 CIE-(Delta) E94 units if the experimental spectral characteristics of a 6-channel display are considered.