Abstract
Abstract: :
Purpose: Observers can readily detect changes in surface color during changes in illuminant (Foster et al., 2001, Proc. Nat. Acad. Sci. USA, 98, 8151-8156). This ability to detect violations of relational color constancy may be based on spatial ratios of cone excitations (Nascimento & Foster, 1997, Proc. Roy. Soc. Lond. B, 264, 1395-1402). The aim here was to test whether this ability depends on the spatial and spectral properties of the scenes used as stimuli. Methods: Simulations of natural scenes and of Mondrian-like colored patterns were presented on a color monitor with 10-bit resolution per gun. The natural scenes were obtained with a fast hyperspectral imaging system (e.g. Nascimento, et al. 2002, J. Opt. Soc., Amer. A, 19, 1484-1490), giving the reflectance function at 10-nm intervals at each point in a high-resolution digital image. Scenes included stones, flowers, and foliage. The Mondrian patterns consisted of 49 (7 × 7) abutting 1.6-deg-square uniform surfaces with spectral reflectances drawn from natural scenes or from the Munsell set. In each trial, two images were presented in sequence, each for 1 s, with no interval. The images differed in illuminant, either a daylight with correlated color temperature 25 000 K and then 6700 K or, depending on spectral content of the natural surfaces, 15 000 K and then 5700 K. The spectral reflectance of a region in the second image was changed randomly in a way quantified by an equivalent local change in daylight. The observer reported whether there was an illuminant change or one with a surface-reflectance change. Results: In terms of mean square error, performance was best with natural scenes, intermediate with Mondrian patterns of natural surfaces, and worst with Mondrian patterns drawn from the Munsell set. Conclusions: Both the spatial characteristics and the spectral content of the natural scenes tested here contribute towards improved color constancy.
Keywords: color appearance/constancy • color vision • scene perception