A Psychophysical Analysis of Illuminant Estimation Algorithms

Abstract

Illuminant estimation algorithms are often evaluated by calculating recovery angular error which is the angle between the RGB of the ground truth and the estimated illuminants. However, the same scene viewed under two different lights with respect to which the same algorithm delivers illuminant estimates and then identical reproductions - and so, the practical estimation error is the same - can, in fact and counterintuitively, result in quite different recovery errors. Reproduction angular error has been recently introduced as an improvement to recovery angular error. The new metric calculates the angle between the RGB values of a white surface corrected by the ground truth illuminant and corrected by the estimated illuminant. Experiments show that illuminant estimation algorithms could be ranked differently depending on whether they are evaluated by recovery or reproduction angular error. In this paper a psychophysical experiment is designed which demonstrates that observers choices on 'what makes a good reproduction' correlates with reproduction error and not recovery error.