In David Ingle's experiment, a goldfish has been trained to swim to a patch of a given color for a reward—a piece of liver. It swims to the green patch regardless of the exact setting of the three projectors' intensities. The behavior is strikingly similar to the perceptual result in humans.
COLOR AND THE SPATIAL VARIABLE
We saw in Chapter 3 that an object's whiteness, blackness, or grayness depends on the light that the object reflects from some source, relative to the light reflected by the other objects in the scene, and that broad-band cells at an early stage in the visual pathway—retinal ganglion cells or geniculate cells—
can go far to account for this perception of black and white and gray: they make just this kind of comparison with their center-surround receptive fields.
This is surely Hering's third, spatially opponent black-white process. That the spatial variable is also important for color first began to be appreciated a century ago. It was tackled analytically only in the last few decades, notably by psychophysicists such as Leo Hurvich and Dorothea Jameson, Deane Judd, and Edwin Land. Land, with a consuming interest in light and photography, was naturally impressed by a camera's failure to compensate for differences in light sources. If a film is balanced so that a picture of a white shirt looks white in tungsten light, the same shirt under a blue sky will be light blue; if the film is manufactured to work in natural light, the shirt under tungsten light will be pink. To take a good color picture we have to take into account not only light intensity, but also the spectral content of the light source, whether it is bluish or reddish. If we have that information, we can set the shutter speed and the lens opening to take care of the intensity and select the film or filters to take care of color balance. Unlike the camera, our visual system does all this automatically, and it solves the problem so well that we are generally not aware that a problem exists. A white shirt thus continues to look white in spite of large shifts in the spectral content of the light source, as in going from overhead sun to setting sun, to tungsten light, or to fluorescent light. The same constancy holds for colored objects, and the phenomenon, as applied to color and white, is called color constancy. Even though color constancy had been recognized for many years, Land's demonstrations in the 1950s came as a great surprise, even to neurophysiologists, physicists, and most psychologists.
What were these demonstrations? In a typical experiment, a patchwork of rectangular papers of various colors resembling a Mondrian painting is illuminated with three slide projectors, one equipped with a red, the second with a green, the third with a blue filter. Each projector is powered by a variable electric source so that its light can be adjusted over a wide range of intensities.
The rest of the room must be completely dark. With all three projectors set at moderate intensities, the colors look much as they do in daylight. The surprising thing is that the exact settings do not seem to matter. Suppose we select a green patch and with a photometer precisely measure the intensity of the light coming from that patch when only one projector is turned on. We then repeat the measurement, first with the second projector and then with the third. That gives us three numbers, representing the light coming to us when we turn on all three projectors. Now we select a different patch, say orange, and readjust each projector's intensity in turn so that the readings we now get from the orange patch are the same as those we got before from the green one. Thus with the three projectors turned on, the composition of light now coming from the orange patch is identical to the composition of light that a moment ago came from the green. What do we expect to see? Naively, we would expect the orange patch to look green. But it still looks orange—indeed, its color has not changed at all. We can repeat this experiment with any two patches. The conclusion is that it doesn't much matter at what intensities the three projectors are set, as long as some light comes from each. In a vivid example of color constancy, we see that twisting the intensity dials for the three projectors to almost any position makes very little difference in the colors of the patches.
Such experiments showed convincingly that the sensation produced in one part of the visual field depends on the light coming from that place and on the light coming from everywhere else in the visual field. Otherwise, how could the same light composition give rise at one time to green and at another to orange? The principle that applies in the domain of black, white, and gray, stated so clearly by Hering, thus applies to color as well. For color, we have an opponency not only locally, in red versus green and yellow versus blue, but also spatially: center red-greenness versus surround red-greenness, and the same opponency for yellow-blueness.
In 1985, in Land's laboratory, David Ingle managed to train goldfish to swim to a patch of some preassigned color in an underwater Mondrian display. He found that a fish goes to the same color, say blue, regardless of wavelength content: it selects a blue patch, as we do, even when the light from it is identical in composition to the light that, in a previous trial and under a different light source, came from a yellow patch, which the fish had rejected.
Thus the fish, too, selects the patch for its color, not for the wavelength content of the light it reflects. This means that the phenomenon of color constancy cannot be regarded as some kind of embellishment recently added by evolution to the color sense of certain higher mammals like ourselves; finding it in a fish suggests that it is a primitive, very basic aspect of color vision. It would be fascinating (and fairly easy) to test and see whether insects with color vision also have the same capability. I would guess that they do.
Land and his group (among others, John McCann, Nigel Daw, Michael Burns, and Hollis Perry) have developed several procedures for predicting the color of an object, given the spectral-energy content of light from each point in the visual field but given no information on the light source. The computation amounts to taking, for each of the three separate projectors, the ratio of the light coming from the spot whose color is to be predicted to the average light coming from the surround. (How much surround should be included has varied in different versions of the theory: in Land's most recent version, the surround effects are assumed to fall off with distance.) The resulting triplet of three numbers—the ratios taken with each projector—uniquely defines the color at that spot. Any color can thus be thought of as corresponding to a point in three-dimensional space whose coordinate axes are the three ratios, taken with red light, green light, and blue light. To make the formulation as realistic as possible, the three lights are chosen to match the spectral sensitivities of the three human cone types.
That color can be so computed predicts color constancy because what counts for each projector are the ratios of light from one region to light from the average surround. The exact intensity settings of the projectors no longer matter: the only stipulation is that we have to have some light from each projector; otherwise no ratio can be taken. One consequence of all this is that to have color at all, we need to have variation in the wavelength content of light across the visual field. We require color borders for color, just as we require luminance borders for black and white. You can easily satisfy yourself that this is true, again using two slide projectors. With a red filter (red cellophane works well) in front of one of the projectors, illuminate any set of objects. My favorite is a white or yellow shirt and a bright red tie. When so lit, neither the shirt nor the tie looks convincingly red: both look pinkish and washed out. Now you illuminate the same combination with the second projector, which is covered with blue cellophane. The shirt looks a washed-out, sickly blue, and the tie looks black: it's a red tie, and red objects don't reflect short wavelengths.
Go back to the red projector, confirming that with it alone, the tie doesn't look especially red. Now add in the blue one. You know that in adding the blue light, you will not get anything more back from the tie—you have just demonstrated that—but when you turn on the blue projector, the red tie suddenly blazes forth with a good bright red. This will convince you that what makes the tie red is not just the light coming to you from the tie.
Experiments with stabilized color borders are consistent with the idea that differences across borders are necessary for color to be seen at all. Alfred Yarbus, whose name came up in the context of eye movements in Chapter 4, showed in 1962 that if you look at a blue patch surrounded by a red background, stabilizing the border of the patch on the retina will cause it to disap
In many of his experiments Edwin Land used a Mondrian-like patchwork of colored papers. The experiments were designed to prove that the colors remain remarkably constant despite marked variations in the relative intensities of the red, green, and blue lights used to illuminate the display.