Info

Although these kinds of results had been demonstrated repeatedly since the nineteenth century, Land's fame and the provocative way he stated his case created a vigorous debate. "We have come to the conclusion," Land opined in a manner that was bound to irritate the vision scientists who had been working on these issues, "that the classic laws of color mixing conceal great basic laws of color vision" (Land, 1959). Much of the ensuing controversy focused on whether Land had shown anything new. He had not, but his demonstrations emphasized that color vision was deeply complicated by the fact that the entire scene is somehow relevant to the perceived color of any part of it (as is true for stimuli that elicit perceptions of lightness or brightness) and that understanding color vision would require understanding these phenomena.

Land tried to explain the interaction of the spectral information in a scene by postulating that contrast and constancy are based on neural computations that entail the ratios of cone activity multiplied by the illuminant at each point in the retinal image. Other investigators more familiar with the physiology of neuronal interactions suggested that the basis of these contextual effects was probably input-level adaptation to the predominant spectral contrasts in a scene. But other stimuli elicit color perceptions that are inconsistent with the predictions of Land's theory or theories based on adaptation. For example, in Figure 9.4, the spectra of the targets (the circles in [A] and the crosses in [B]) are identical in all four panels, and the spectra of the orange and yellowish backgrounds in the upper and lower panels are also the same. However, the targets on the left in both the upper and lower panels appear to be about the same color, as do the targets on the right. These similar color perceptions of the targets occur despite the fact that the surrounds are opposite in the upper and lower panels. This confound is a similar to the puzzle presented by the grayscale stimulus in Figure 8.6.

Figure 9.4 A further challenge to rationalizing color contrast and constancy. A) A typical color contrast stimulus in which two identical central targets embedded in different spectral surrounds appear differently colored (see Figure 9.2). B) In this configuration, the central targets are physically the same as those in the upper panels (see the key), but the backgrounds have been switched. Despite this switch, the apparent colors of the left and right targets are about the same in the upper and lower panels. (From Purves and Lotto, 2003)

This bit of history provides some sense of the challenges facing anyone who wants to say something sensible about why we see the colors we do. But based on the inverse problem and the accounts that we had been able to give for otherwise puzzling brightness effects such as Mach bands and the Cornsweet edge effect (see Chapter 8), Lotto and I felt pretty sure that color perceptions must also be explainable in empirical terms. After all, the entanglement of the factors generating luminance values illustrated in Figure 8.4 applies equally to spectral information: The conflation of the illuminant, the surface properties of objects, and the subtraction of light by the atmosphere would make the real-world sources of color spectra just as uncertain as the sources of the overall amount of light in visual stimuli. Because no logical operation on the distribution of light energy in retinal images—the basis of color vision—could specify what real-world objects and conditions had given rise to the relevant stimulus, color perceptions would also have to be determined by accumulated trial and error experience.

For a number of reasons—including the contentious nature of the subject and its inherent complexity—developing an empirical theory of color vision is considerably more challenging than rationalizing the relationship between luminance and lightness/brightness in these terms. Nevertheless we began exploring the idea that the odd way we see color is the result of trial-and-error experience with spectral relationships accumulated over evolutionary time. Our supposition was that color contrast and constancy, similar to the lightness/brightness phenomena described in Chapter 8, must arise from linking spectral stimuli and physical sources according to behaviors that worked in the past. If so, then the nature of this accumulated empirical information should predict color perceptions.

We started out testing this idea by having subjects adjust the perceived color of a target on a neutral background until it matched an identical target on a color background, measuring in this way the perceptual change induced by the color of the surround with stimuli similar to the example in Figure 9.2. The relationships that we determined in this way did not rule out Land's theory, adaptation, or other schemes based on interactions among neurons sensitive to color stimuli. But they did show that color contrast could be equally well understood as the outcome of a visual strategy in which color perceptions are generated according to past experience. Much of this work depended on

Lotto's extraordinary skill as an artist, and Figure 9.5 shows him with one of the public installations that he produced to demonstrate the ideas described here.

Figure 9.5 Beau Lotto in 2008 with one of his public art installations in London. (Courtesy of Beau Lotto)

Figure 9.6 shows how an empirical explanation of color perception would work in the case of a color contrast stimulus. The genesis of the light coming from the targets in the standard color contrast stimulus in Figure 9.6A is not knowable directly; as indicated in the cartoons in Figures 9.6B and 9.6C, many different combinations of reflectance, illumination, and transmittance could have produced the same stimulus. Accordingly, the biological strategy outlined in Chapter 8 for determining what lightness or brightness an observer should see in response to a given luminance (see Figure 8.7) is equally applicable to color: Contend with the inverse problem by color perceptions based on neural circuitry determined by operationally successful behavior in the past. The circuitry needed to facilitate successful behavior would have been shaped according to on how often the stimulus in Figure 9.6 A was generated by the situation in Figure 9.6B versus 9.6C, or any of the many other combinations of surface properties, illumination, and other factors that could produce the stimulus in Figure 9.6A. The different color appearance of the two targets would be the perceptual signature of this way of contending with the inverse problem.

Figure 9.6 An empirical explanation of color contrast. The standard color contrast stimulus in (A) could have been generated by physically identical targets on physically different backgrounds under white light illumination (as in [B]), or physically different surfaces on physically similar backgrounds under illuminants with different color spectra (as in [C]). In an empirical framework, the different color appearance of the identical targets in (A) is a result of operational links made over evolutionary and individual time between the inevitably uncertain meaning of the spectral characteristics of the retinal image and experience with the relative success of behavioral responses made to the same or similar stimuli in the past. (After Purves, et al., 2001)

ivi

Beau and I gained more confidence in this explanation of color perception by manipulating spectral patterns that enhanced or diminished color contrast by making the characteristics of a scene either more or less consistent with different possible bases for the light coming from target patches ( Figure 9.7). For example, when identical targets (the central squares) are presented on backgrounds that comprise a variety of tiles with spectra designed so that the two arrays are likely to be under "red" and "blue" illumination, respectively, the apparent color difference between the targets is relatively marked ( Figure 9.7A). Conversely, when Lotto made the contextual information consistent with the arrays under the same illumination, the apparent color difference of the physically identical targets decreased ( Figure 9.7B). Because the average spectral content in the left and right scenes is the same, these effects cannot be explained by neuronal interactions at the input stages of the visual system or adaptation. However, in empirical terms, the color perceptions elicited by the same targets in different contexts make good sense. The waxing of circuitry underlying successful behaviors over time would have incorporated the empirical fact that when the light spectrum coming from two patches is the same but the illumination is different, the objects are always physically different surfaces. However, when the light from the patches is the same and they are under the same illumination, experience would have taught that the objects are physically the same. The rationale for the differences we actually see (color contrast effects) is that two things that are different must look different for behavior to be successful, even when they are returning exactly the same light spectrum to the eye.

The same argument can explain color constancy in empirical terms. Most people had assumed (and many still do) that color constancy is an explicit goal of vision, meaning that bananas, apples, or any other object should continue to look their respective colors in all circumstances so that we can more readily identify the objects (the usual example given is identifying the ripeness of fruits, an ability that would presumably have evolutionary value). We thought it was more likely that color constancy is just another indicator that the colors we see are generated empirically, with color contrast and constancy being two sides of the same coin. If the information in a scene is consistent with accumulated experience of interactions with two targets that return the same spectra to the eye but turn out to be different physical surfaces, then they will appear differently colored (color contrast); conversely, if the information in a scene is inconsistent with this possibility, then they will tend look similarly colored, even if the two targets return different spectra to the eye (color constancy).

Figure 9.7 Altering color perception by manipulation of empirical significance. A) A stimulus in which a physically identical central tile is presented in the context of other tiles whose spectral returns are all consistent with "reddish" illumination of the left panel and "bluish" illumination of the right panel. B) The same scene as (A), but with the central target surrounded by tiles whose spectral returns are consistent with similar illumination of both left and right panels. See text for explanation. (After Lotto and Purves, 2000. Copyright 2000 National Academy of Sciences, U.S.A.)

Figure 9.8 sums up this interpretation of color contrast and constancy. Although the spectra coming from the central squares on the two faces of the cube in Figure 9.8A are identical, their colors are different because the information in the stimulus is what we humans would have experienced when interacting with differently reflective surfaces under different illuminants. Conversely, the same two targets on the faces of the cube in Figure 9.8B appear relatively similar even though the spectra coming from the targets are different, because this is the information we would always have experienced when interacting with physically similar surfaces under different illumination. In both instances, seeing color in this way would have fomented successful behavior.

Figure 9.8 Color contrast and constancy as consequences of the same empirical strategy of color vision. The two panels demonstrate the effects on the color appearance of surfaces when two physically similar target surfaces (as in [A]) or two physically different target surfaces (as in [B]) are presented in the same context. Because the spectral information in the scene is consistent with different intensities of light falling on the upper and lateral surfaces of the cube, a color contrast effect is seen in (A) and a color constancy effect is seen in (B). The appearances of the relevant target surfaces in the same neutral context are shown in the keys. (From Purves and Lotto, 2003)

Was this article helpful?

0 0

Post a comment