Colour constancy using von Kries transformations: colour constancy "goes to the Lab"

Abstract

Colour constancy algorithms aim at correcting colour towards a correct perception within scenes. To achieve this goal they estimate a white point (the illuminant's colour), and correct the scene for its in uence. In contrast, colour management performs on input images colour transformations according to a pre-established input pro le (ICC pro le) for the given con- stellation of input device (camera) and conditions (illumination situation). The latter case presents a much more analytic approach (it is not based on an estimation), and is based on solid colour science and current industry best practises, but it is rather in exible towards cases with altered conditions or capturing devices. The idea as outlined in this paper is to take up the idea of working on visually linearised and device independent CIE colour spaces as used in colour management, and to try to apply them in the eld of colour constancy. For this purpose two of the most well known colour constancy algorithms White Patch Retinex and Grey World Assumption have been ported to also work on colours in the CIE LAB colour space. Barnard's popular benchmarking set of imagery was corrected with the original imple- mentations as a reference and the modi ed algorithms. The results appeared to be promising, but they also revealed strengths and weaknesses.