Abstract
Observers can readily discriminate between illuminant changes and surface-color changes in images of natural scenes, but performance varies with scene content (Amano et al., 2004, Perception, 33, Suppl. 65), and, in particular, may be affected by the color of a test surface used to probe surface-color judgments. To address this issue, test-surface color was varied systematically in a discrimination experiment. Images of natural rural and urban scenes were presented on a computer-controlled color monitor with 10-bit resolution per gun. The scenes were reproduced from data obtained with a high-resolution hyperspectral imaging system (Foster et al., 2004, Visual Neurosci., 21, 331–336). Test-surface color was manipulated so that, in separate sessions, it was initially neutral, reddish, bluish, greenish, and yellowish. In each trial, two images of a pattern were presented in sequence, each for 1 s with no interval, under two different daylight illuminants: the first with correlated color temperature 25000 K or 4000 K, the second 6700 K. The spectral reflectance of the test surface in the second image was changed randomly from trial to trial. The images, viewed at 100 cm, subtended approx. 17° × 14°. Observers had to decide whether the test surface in the successive images was the same. Performance by 8 observers with normal color-vision was quantified with a standard constancy index (Arend et al., 1991, J. Opt. Soc. Am. A, 8, 661–672). Performance was generally high, but varied over scenes from 0.40 to 0.95 (with 1.0 representing ideal performance). Test color interacted significantly with direction of illuminant change (F(4,64)=3.8, p =0.01). Interestingly, a neutral test surface did not give best performance uniformly across scenes (Nascimento et al., 2004, Visual Neurosci. 21, 337–340), suggesting a more complex interaction between surface colors in determining surface-color perception in natural scenes.