Colour matching function learning

Abstract

In this paper, we aim at learning the colour matching functions making use of hyperspectral and trichromatic imagery. The method presented here is quite general in nature, being data driven and devoid of constrained setups. Here, we adopt a probabilistic formulation so as to recover the colour matching functions directly from trichromatic and hyperspectral pixel pairs. To do this, we derive a log-likelihood function which is governed by both, the spectra-to-colour equivalence and a generative model for the colour matching functions. Cast into a probabilistic setting, we employ the EM algorithm for purposes of maximum a posteriori inference, where the M-step is effected making use of Levenberg-Marquardt optimisation. We present results on real-world data and provide a quantitative analysis based upon a colour calibration chart. © 2012 Springer-Verlag Berlin Heidelberg.