Abstract
PURPOSE. Some studies of lightness and color constancy have described asymmetric matching conditions under which observers are unable to find a perfect match; that is, the best match was associated with a residual perceptual difference (e.g., Brainard, Brunt, and Speigle, 1997). Logvinenko and Maloney (2006) quantified this phenomenon by having observers rate differences in perceived surface lightness across changes surface reflectance and illumination, and modeling the dissimilarity data using multidimensional scaling (MDS). Their data required a two-dimensional representation for lightness, with one dimension corresponding (roughly) to surface reflectance and the other (roughly) to illumination intensity. The finding of a two-dimensional representation explains why observers cannot make a perfect asymmetric lightness match across a change of illumination by adjusting the match stimulus intensity alone. We sought to replicate Logvinenko and Maloney's result, and to extend it to the case where viewing geometry was also varied. METHODS. Observers viewed pairs of grayscale matte flat test stimuli, one presented in each of two adjacent illuminated viewing chambers. They rated the dissimilarity of each pair. Observers scaled all possible pairs from a stimulus set containing 6 surface reflectances seen in 3 scene contexts. Across one context change illumination varied, while across the other surface slant varied. Non-metric MDS was used to model the data. RESULTS. The data from each of four observers were well-accounted for by a one-dimensional representation. This representation was similar in structure for all observers; for each the position p of a surface with reflectance r was approximated by p = log(r) + c, where the constant c depended on scene context. Accounting for our dissimilarity data did not require two or more dimensions of perceived lightness. CONCLUSION. Understanding what variations of viewing conditions produce a lightness representation with multiple dimensions remains to be determined.
Supported by NIH RO1 EY10016 and P30 EY001583.