Professional Documents
Culture Documents
context effects
Fuhui Long* and Dale Purves
Center for Cognitive Neuroscience and Department of Neurobiology, Box 90999, Levine Science Research Center, Room B243E, Research Drive,
Duke University, Durham, NC 27708
The color context effects referred to as color contrast, constancy, measured the radiance of each image pixel as a function of
and assimilation underscore the fact that color percepts do not wavelength (3–5). These images were obtained with the intention
correspond to the spectral characteristics of the generative stimuli. of representing typical natural environments and have been used
Despite a variety of proposed theories, these phenomena have previously to investigate the distributions of spectral returns,
resisted explanation in a single principled framework. Using a reflectances, and illuminants (4–5).
hyperspectral image database of natural scenes, we here show Rectangular regions of fixed size (typically 10% of the overall
that color contrast, constancy, and assimilation are all predicted by image size) were randomly sampled in all of the images in the
the statistical organization of spectral returns from natural visual database and analyzed in terms of target and context (Fig. 2). To
environments. obtain samples pertinent to color contrast, constancy, and
assimilation stimuli by using the same method, the target was
defined by spatially isolated patches of the same size (1.5% of the
I t has long been appreciated that color percepts, like other
perceived visual qualities, do not correspond in any simple way
to the distribution of light energy in spectral returns. Thus, two
region sampled) and value in the 1931 Commission Internatio-
nale de L’Eclairage (CIE) chromaticity diagram (6) (Fig. 2 B2
visual targets generating identical spectra can, as a consequence and C2); the context was defined as the remaining area. When
of context, elicit different color percepts (Fig. 1 A and C), the number of target and contextual elements was relatively
whereas different spectral returns can elicit more similar colors small (e.g., ⬍10), the samples approximated the configuration of
(Fig. 1 B and D). In color contrast and constancy stimuli, the color contrast and constancy stimuli (see Fig. 1 A and B); when,
colors generated by the targets are shifted away from the colors however, the number of the patches was relatively large (e.g.,
elicited by their respective surrounds (Fig. 1 A and B, respec- ⬎20), the samples more nearly approximated the configuration
tively). On the other hand, in color assimilation stimuli (Fig. 1 C of color assimilation stimuli (see Fig. 1 C and D).
and D), highly repetitive spatial patterns cause the apparent Separating each sampled region into target and context was
colors of the patterned elements to be shifted toward each other. carried out in two steps. First, a small patch with a relatively
The variety and complexity of these contextual effects have uniform spectrum (SD of the chromaticity of all of the pixels in
made it extraordinarily difficult to rationalize color perception in the patch divided by their mean chromaticity ⫽ ⬍0.03) was
any of the several conceptual frameworks that have been pro- selected near the center of the region being analyzed (see boxed
posed (see Discussion). Here, we examine the idea that color patch in Fig. 2 B2 and C2). Using this patch as a reference, the
contrast, constancy, and assimilation are all generated by past remaining area of the region was sampled to identify all of the
experience with the statistical organization of spectral returns other patches of similar chromaticity. To be accepted as a
arising from natural visual environments. The biological ratio- component of the target (see definition above), any given patch
nale underlying this idea is that, because of the conflation of the had to be at least one pixel away from any other patch, both
generative sources of visual stimuli in the retinal image, spectral horizontally and vertically. The set of all such patches was then
stimuli are inherently ambiguous (1, 2). Accordingly, if visually defined as the target in the sampled region, as illustrated by the
guided behavior is to be successful, visual percepts must in some patches on the black background in Fig. 2 B2 and C2. In a second
way be determined by the statistical organization of spectral step, the context of the target was identified in each sample. If
returns in natural environments that humans have typically the chromaticity of the area remaining in the region sampled was
encountered, rather than by the physical properties of the relatively uniform, it was accepted as the context of the target
particular stimulus giving rise to the retinal image (2). If this (see Fig. 2 B3 and C3). If the remaining area failed to meet this
supposition is correct, the perceived color of a target in any given criterion, the sample was excluded from further analysis and the
stimulus should be predicted by the typical chromaticity of the procedure was repeated beginning with another region in the
target in the chromatic and spatial contexts that occur in natural image. Although the set of target and context areas identified in
scenes. this way in natural images is inevitably less uniform than in the
This hypothesis was tested by analyzing the co-occurrence of ‘‘laboratory’’ stimuli illustrated in Fig. 1, this approach identifies
targets and contexts in a database of hyperspectral images of the relevant stimulus categories as they occur in normal viewing
natural scenes. A large number of samples was collected by using (see Discussion for further explanation of the underlying ratio-
criteria that mimicked the spatial complexity and contextual nale of this approach). In this way, we generated a total of ⬇1.5
chromaticity of the standard color contrast, constancy, and million samples that approximated, as closely as possible in
assimilation stimuli illustrated in Fig. 1 (see Fig. 2 and Methods). natural images, the color contrast, constancy, and assimilation
In this way, we could evaluate the conditional probability stimuli that have generally been used in psychophysical studies.
distribution of the spectral characteristics of a target, given the
spectral characteristics of the context and the spatial complexity Statistical Analysis. We next computed the chromaticity of these
of the sample. We then asked whether the perceived color of the components (denoted Ct and Cc) by averaging the values of all
target in such stimuli is predicted by these distributions.
Long and Purves PNAS 兩 December 9, 2003 兩 vol. 100 兩 no. 25 兩 15191
Fig. 4. Average chromaticity difference between target and context (兩Cc-Ct兩)
as a function of the number of the iterated elements in the sampled targets
(Nt).
NEUROSCIENCE
it is not difficult to create stimuli that elicit color perceptions fact that the typical chromaticity of a target in natural scenes is
opposite that of the context, and that it tends to change to a value
opposite those predicted by theories based on lateral integration
similar to that of the context as the spatial complexity of the
(17–18).
sample increases, can be understood intuitively. When only one
Another attempt to rationalize color context effects has
or a few target elements are identified in a spectrally homoge-
appealed to multiple spatial frequency filtering mechanisms
nous surround (as in a color contrast and constancy stimuli), the
(19–23). In this scenario, color contrast is taken to be generated
real-world sources of target and context tend to be physically
by neuronal responses tuned to low spatial frequency stimuli and different surfaces in the same illumination; on the other hand,
assimilation by neuronal responses tuned to both low and high when many target elements are found in a spectrally homoge-
spatial frequency components. Although this further mechanism nous surround, the sources of the target and context are likely be
is helpful in accounting for some color context effects, color physically similar surfaces. (This does not mean, of course, that
perceptual phenomena that depend on scene properties such the target and context in any stimulus with high spatial com-
as orientation and depth also resist this sort of explanatory plexity will be perceived to be the same color; recall that in this
framework. framework the perceived color of a target is determined by how
Given these uncertainties and the implications of other recent the position of a spectral return is shifted in subjective color
work on color perception (2, 24) the hypothesis that color space by the corresponding spectral contexts and spatial com-
contrast, constancy, and assimilation effects (and by inference all plexity that humans have typically encountered.) Because any
color percepts) are generated by the statistical organization of visual stimulus conflates its generative physical sources (e.g.,
the spectral returns arising from natural visual environments is reflectance and illumination), these systematic changes in the
an attractive one. The present analysis of target and context probable physical sources conveyed by the statistical organiza-
samples that approximate standard color contrast, constancy, tion of the spectral returns in natural scenes will systematically
and assimilation stimuli show that the probability distribution of change the perceptual effects of inherently ambiguous stimuli.
the spectral characteristics of targets in nature vary systemati-
cally as a function of the spectral characteristics of the context We thank Catherine Howe, Shuro Nundy, David Schwartz, James
and its spatial complexity. Thus, the probability distributions of Voyvodic, and Zhiyong Yang for helpful criticism.
1. Marr, D. (1982) Vision: A Computational Investigation into Human Represen- 9. Judd, D. B. (1940) J. Opt. Soc. Am. A 30, 2–32.
tation and Processing of Visual Information (Freeman, San Francisco). 10. Land, E. H. & MaCann, J. J. (1971) J. Opt. Soc. Am. A 61, 1–11.
2. Purves, D. & Lotto, R. B. (2003) Why We See What We Do: An Empirical Theory 11. Hurvich, L. M. (1981) Color Vision (Sinauer, Sunderland, MA)
of Vision (Sinauer, Sunderland, MA). 12. Land, E. H. (1986) Vision Res. 26, 7–21.
3. Brelstaff, G., Parraga, A., Troscianko, T. & Carr, D. (1995) in Proceedings of 13. Benary, W. (1924) Psychol. Forsch. 5, 131–142.
SPIE 2587: Geographic Information Systems, Photogrammetry, and Geological兾 14. Knill, D. & Kersten, D. (1991) Nature 351, 228–230.
Geophysical Remote Sensing, eds. Lurie, J. B., Pearson, J. J. & Zillioli, E. 15. Adelson, E. H. (1993) Science 262, 2042–2044.
(Internatl. Soc. Opt. Eng., Bellingham, WA), pp. 150–159. 16. Brown, R. O. & MacLeod, D. I. (1997) Curr. Biol. 7, 844–849.
4. Parraga, C. A., Brelstaff, G. J., Troscianko, T. & Moorhead, I. (1998) J. Opt. 17. White, M. (1979) Perception 8, 413–416.
Soc. Am. A 15, 563–569. 18. Todorovic, D. (1997) Perception 26, 379–394.
5. Chiao, C. C., Cronin, T. W. & Osorio, D. (2000) J. Opt. Soc. Am. A 7, 218–224. 19. Blaskeslee, B. & McCourt, M. E. (1997) Vision Res. 37, 2849–2869.
6. Commission Internationale de L’Eclairage (1986) Colorimetry (Central Bureau 20. McCourt, M. E. (1982) Vision Res. 22, 119–134.
of the Commission Internationale de L’Eclairage, Vienna), 2nd Ed., Publ. 15.2. 21. Zaidi, Q. (1989) Vision Res. 29, 691–697.
7. Hering, E. (1920) Grundzüge der Lehre von Lichtsinn (Springer, Berlin), trans. 22. DeValois, R. L. & DeValois, K. K. (1988) Spatial Vision (Oxford Univ. Press,
Hurvich, L. M. & Jameson, D. J. (1964) Outlines of a Theory of the Light Sense New York).
(Harvard Univ. Press, Cambridge, MA). 23. Moulden, B. & Kingdom, F. A. (1991) Vision Res. 31, 1999–2008.
8. Helson, H. (1938) J. Exp. Psychol. 23, 439–471. 24. Lotto, R. B. & Purves, D. (2000) Proc. Natl. Acad. Sci. USA 97, 12834–12839.
Long and Purves PNAS 兩 December 9, 2003 兩 vol. 100 兩 no. 25 兩 15193