Abstract
Colour constancy is the ability of the human visual system to keep object colours roughly constant over changes in illumination. Colour constancy measurements are typically made for a limited number of illumination conditions, and often for simulated rather than real-world scenes. The generalisability of the results is therefore unclear. Here we measure colour constancy for a scene with real objects under a broad range of illuminations, both on and off the daylight locus. The setup consists of a spectrally tunable LED illuminator that emits diffuse, nearly uniform light into an enclosed box of approximately 42 x 60 x 75cm (hwd). Subjects viewed the interior of the box through a viewport, and adjusted the chromaticity of an LCD-display patch visible through a cut-out in the Mondrian paper lining the box, using a mobile black keypad reached through an armhole in the box front. Their task was to adjust the patch to appear grey, under each of eight distinct smooth broad-band test illuminations whose chromaticity coordinates spanned an ellipse in Lab space centred on daylight D65 (at distances 70 – 120 ΔE). Subjects initially dark adapted for 5 minutes. On each trial, subjects adapted to D65 for 10 seconds, followed by adaptation to the test illumination for 10 seconds during which they adjusted the patch chromaticity. The cycle repeated until the subject signalled a match to "grey". On average, colour constancy was imperfect: the adjusted "grey" chromaticity deviated significantly from the illumination chromaticity, but was significantly better for test illumination chromaticities on the daylight locus. For test chromaticities off the daylight locus, errors were lower for "redder" and "yellower" test illuminations. Colour constancy thus varies across illuminations, and may be better adapted to "warm" illuminations.
Meeting abstract presented at VSS 2012