Abstract
Colour constancy measurements typically involve the participant viewing a limited-field scene (e.g., on a display screen) from an external vantage point (e.g., through a porthole) and comparing surface colours across a small number of distinct illuminations. Here we report colour constancy measurements from an immersive viewing setup in which scene appearance comparisons are made between multiple distinct illuminations. The setup consists of an enclosed room (approx. 2.5m x 2.5m x 2.5m) with all interior surfaces painted white, illuminated solely by multiple-channel LED light sources whose outputs are spectrally tuneable in real-time. Participants (n=6) sat inside the room on a black bench, facing a Mondrian-paper-lined three-sided open box positioned on a black pedestal approx. 1.5m distant, and gave responses via a black joystick. There were no other room contents. A 2-down, 1-up staircase protocol determined just-discriminable global illumination changes. Poorer discrimination of illumination change implies better colour constancy. On each trial, the reference illumination (D67) was shown for 2000ms followed by two test illuminations of 500ms each; each illumination change was preceded by a 400ms dark period. One test illumination was identical to the reference; the chromaticity of the other varied parametrically in uniform colour space from the reference target in one of four chromatic directions, blue or yellow (on the daylight locus), or red or green (on an orthogonal locus). All illuminations were the smoothest-possible metamers for the requested chromaticity. Illumination discrimination thresholds, calculated by averaging final staircase reversals, were lower on the orthogonal red-green locus than the daylight locus (p <0.06), and significantly higher for "bluer" illumination changes than "greener" directions (∆Euv = 17.2 vs 6.2; p <0.03). Global illumination change discrimination thus provides a robust method for assessing colour constancy in immersive viewing, and the results support the notion that constancy mechanisms are biased towards daylight illuminations.
Meeting abstract presented at VSS 2014