Abstract
In everyday life, surface colours typically appear constant under different lighting despite the (potentially) dramatic changes in luminance—human observers are (nearly) colour constant. Failures of colour constancy using simple artificial stimuli are often explicable in terms of luminance normalization and color adaptation, but not in realistic scenes with multiple light sources in which the perceiver ought to separate all illuminants from the surface reflectances. What remains to be established, however, is the link between lightness constancy using simple artificial stimuli versus realistic scenes. Here we present a lightness scaling experiment to investigate the impact of realism on lightness perception. Observers see physically accurate renderings of indoor scenes that contain three target patches of varying illumination and luminance; the observers have to judge which lateral target patch appears more similar to the center target. Based on these triad judgments, we estimated 1D and 2D lightness scales for different manipulations of scene realism with an ordinal embedding algorithm, a modern variant of multidimensional scaling. Preliminary results show a mixed picture of lightness constancy's dependence on scene realism. Scene realism per se appears not critical; instead, we observe lightness constancy if the scene contains spotlights with clearly visible, sharp light cones. Targets in a scene without sharp light cones—but still clearly visible illuminants, e.g. a ceiling window—were judged based on the brightness both in the full and reduced realism conditions. Our results may indicate that it is not realism per se that is crucial for lightness constancy but the ease with which individual illuminants be identified through the local context.