The requirement that the sensor responses of a camera to a,given surface re
flectance be constant under changing illumination conditions has led to the
development of the so called colour constancy algorithms. Given an image r
ecorded under an unknown illuminant, the task for a colour constancy algori
thm is to recover an estimate of the scene illuminant. One such algorithm d
eveloped by D.A. Forsyth, A novel algorithm for colour constancy, Internati
onal Journal of Computer Vision 5 (1) (1990) 5-36 [1] and later extended by
G.D. Finlayson, Color in perspective, IEEE Transactions on Pattern Analysi
s and Machine Intelligence 18(10) (1996) 1034-1038 [2] exploits the constra
int that under a canonical illuminant all surface colours fall within a max
imal convex set-the canonical gamut. Given a set of image colours Forsyth's
algorithm recovers the set of mappings which cake these colours into the c
anonical gamut. This feasible set of mappings represents all illuminants, w
hich are consistent with the recorded image colours. In this article we add
ress the question of how best to select a single mapping from this feasible
set as an estimate of the unknown illuminant.
We develop our approach in the context of Finlayson's colour-in-perspective
algorithm. This algorithm performs a perspective transform on the sensor d
ata to discard intensity information which, without unrealistic constraints
(uniform illumination and no specularities) being placed on the world, can
not be recovered accurately. Unfortunately, the feasible set of mappings re
covered by this algorithm is also perspectively distorted. Here, we argue t
hat this distortion must be removed prior to carving out map selection and
show that this is easily achieved by inverting the perspective transform. A
mean-selection criterion operating on non-perspective mapping space provid
es good colour constancy for a variety of synthetic and real images. Import
antly, constancy performance surpasses all other existing methods. (C) 1999
Elsevier Science B.V. All rights reserved.