Abstract
Computational models predict that colour constancy should improve with increasing numbers of distinct surface reflectances in a scene. The presence of familiar objects may improve constancy by providing reference surfaces for calibration. Colour constancy may also be optimal for daylight illuminations in which it evolved. To probe these potential influences, we have developed a colour constancy measurement that does not require explicit colour matching between objects or achromatic surface adjustment. Here we measure colour constancy by matching illuminations. We used a tuneable illuminator that emits diffuse, nearly uniform light into an enclosed grey box (h35,w65,d55 cm) containing either a colour checker chart or a banana; participants viewed the scene through a viewport. Target illuminants were taken from the daylight locus and an orthogonal locus with the same correlated colour temperature (CCT). Participants (n = 6) were asked to adjust the illumination in ‘warmer’ or ‘cooler’ directions by using two keys on mobile, black keypad, placed on the box floor; keypresses changed the chromaticity in steps of 1 ΔE along one of the loci, depending on trial block. Participants initially dark adapted for 5 minutes. Each trial began with 10 seconds exposure to the target illumination. A sound signalled an illumination change (to a random chromaticity on the designated locus) and the start of the (untimed) adjustment phase. Accuracy was measured as end-distance from the target illumination in perceptually uniform space. For adjustments along the orthogonal locus, accuracy was higher in the presence of the colour checker chart than the banana, (p <.05). In the presence of the checker, higher accuracy was obtained along the orthogonal than the daylight locus (p = .05). Colour constancy corrections thus seem biased towards the daylight locus, and off-daylight-locus illuminations are corrected for more effectively through multiple reflectances than single familiar objects.
Meeting abstract presented at VSS 2012