Abstract
We conducted a series of experiments to assess how the complexity of surface mesostructure and reflectance properties influence lightness judgments, and how these judgments provide insight into the computations and representations responsible for the perception of lightness in different classes of centre-surround displays. Methods. Flat, matte test patches embedded in 3D rocky surrounds (matte or glossy) were rendered in a natural illumination field. Experiment 1 evaluated the role of surface relief and gloss on perceived lightness, which was compared to two control displays that contained matched pixel histogram or an equated (phase-scramble) power spectrum. Experiment 2 evaluated whether transparency plays a role in the "crispening effect" observed with the homogeneous surrounds. Observers adjusted the lightness and transmittance of a match patch, but were given either mid-level (transparency) or low-level (contrast) task instructions. Experiment 3 used the matches from the second experiment to create stimuli for a 2AFC paired comparison task where observers selected the stimuli which best matched the lightness of the uniform displays. Results. Experiment 1 showed that both surface relief and gloss improved lightness constancy, but the results were indistinguishable from the displays with equated power spectrums. Experiment 2 revealed that observers only varied transmittance when instructed to equate match patch contrast, not when directed to equate transparency and lightness. The results of Experiment 3 revealed the transparent match pattern under the contrast matching instructions generated the best overall lightness matches. Conclusions. For complex displays, the effects of surface mesostructure and surface optics are well explained by "low level" distributions of contrasts across space and scale. Homogeneous displays are paradoxically complex: although they are best matched with transparent stimuli, observers do not appear to have direct phenomenal access to the qualities of the transparent representation, and can only achieve these matches by relying on local contrast cues.
Meeting abstract presented at VSS 2014