Color Vision
Peter Gouras
[Introduction] [The Evolution of Vision] [Color Vision] [Chromatic versus Achromatic Contrast] [Divariant Mammalian Color Vision] [Simultaneous Contrast] [Trivariant Human Color Vision] [Hue, Saturation and Brightness] [The Hering Theory of Color Vision] [Hering Red-Green Channel in the Retina] [Hering Blue-Yellow Channel in the Retina] [Hering White-Black Channel in the Retina] [Retinal Interneurons] [The Role of Phasic Ganglion Cells] [The Lateral Geniculate Nucleus] [Color Vision in Visual Cortex] [Color Vision beyond Striate Cortex] [Color and Form] [References]
1. Introduction.
 Fig. 1. Geometric drawing with 15 or more colors (59 K jpeg image) |
 Fig. 2. Geometric drawing in black and white and shades of gray (59 K jpeg image) |
Color vision is known best by man's perception of it. It creates a unique dimension to sight that is impossible to appreciate by any non-visual means. It depends on wavelength more than on the energy of light but it is an illusion of reality resulting from a comparison of the responses of nerve cells in our brain. Color and all vision are in a sense illusory depending only on messages that pass between millions of neurons that reside within the darkness of our skull. These visual messages allow us to project ourselves into a universe that would be unknown to us without vision.
Much is known about human color vision both subjectively and quantitatively from the fields of physics, psychology and physiology. Physiology attempts to explain color vision by the responses of neurons. This is the ultimate step in understanding color and eventually perhaps in constructing machines that will see a similar universe of colors.
Fig. 3. The visual pathways from retina to visual cortex of the human brain (98 K jpeg image)
Linking subjective human experience with single neural responses, however, has many pitfalls. Most of the information comes from electrophysiological recordings from neurons at the peripheral levels of vision, the retina and the lateral geniculate nucleus of monkeys, while subjective human experience involves the entire brain (Fig. 3). There are vast areas of the visual brain, especially cerebral cortex, that remain relatively uncharted because they are so unresponsive in anesthetized monkeys. But it is only in these centers that the subjective experience of color occurs. Therefore, we have to use our imagination to extrapolate from the early stages in vision at the retinal level to explain the physiology of color vision at higher levels in our brain. This is where the major difficulty lies. It will ultimately be resolved by new techniques that allow examination of the awake, perceiving human brain.
I start by presenting my ideas on how human color vision has evolved from a primative divariant system and go on to consider the transmission of hue, form and contrast in parallel streams from the retina to the cerebral cortex. I suggest neural circuitry underlying the perception of color in the retina, lateral geniculate and striate cortex and beyond.
2. The Evolution of Vision
Vision must have started with organisms detecting the difference between light and dark. Molecular genetics reveals that this began in broad daylight with cone photoreceptors (Bowmaker, 1998), a hypothesis suggested earlier by the comparative anatomist, Gordon Walls (1942). Under these conditions shadows must have been the main stimuli for detecting movements and objects. A shadow depolarizes, i.e. activates, a cone, which releases a neurotransmitter that depolarizes second order retinal neurons, bipolar cells, called off-center bipolars and horizontal cells. The reappearance of light by movement hyperpolarizes the cones. This stops the release of the transmitter, which in turn stops, i.e. hyperpolarizes, the off-bipolars and horizontal cells. This also disinhibits a parallel system of bipolars, called on-bipolars. Disinhibition occurs because the same transmitter released by the cone inhibits the on- and excites the off-bipolars. On- and off- cone bipolars have different receptors for the same transmitter (see chapter on the Outer plexiform layer). This push-pull arrangement of on and off -bipolars (Fig. 4) provides the main input to the brain about the visual universe. One channel signals darkness; the other signals lightness.
Fig. 4. The push-pull arrangement of on- and off-bipolar cells (59 K jpeg image)
In order for cones to function under the enormous changes in ambient light in sunlit scenes, they must regulate their sensitivity. This has evolved within the biochemistry underlying phototransduction (Fig. 5). Light absorbed by a protein (opsin from the Greek word to see) coupled to the 11-cis isomer of retinaldehyde activates the opsin, which in turn activates another protein, transducin. This in turn activates an enzyme, phosphodiesterase that breaks down a cyclic nucleotide to close the cone's membrane to depolarizing ions. This chain of reactions amplifies the light signal. A shadow reverses this process opening the membrane to depolarizing ions. The brighter the ambient light, the briefer is the phototransduction reaction and the faster is it reversed by a shadow. This increased speed of the amplification process reduces sensitivity but quickens the response. This is an important part of visual adaptation, which allows cones to function over a large range of light energy. In addition, there is also negative feedback mediated by horizontal cells that antagonize the responses of cones.
Fig. 5. The phototransduction cascade in rod or cone outer segments (59 K jpeg image)
In a world seen by only one type of cone, objects appear lighter or darker but not in color. Whether an object is light or dark depends on a comparison between the light energy it reflects and that reflected by its background. The brighter the background, the darker the object appears. The darker the background, the lighter the object appears. Absolute energy is sacrificed for relative energy, which is all that is necessary for survival.
3. Color Vision
Inhabiting a world in which an organism can only distinguish light and dark without color is a handicap. Therefore early in the evolution of vision, color must have appeared. For this, two different types of cones were necessary, one responding best to one part and a second responding best to another part of the visible spectrum, i.e. sunlight. By this means the brain can compare two signals to distinguish color. Again the brain senses relative rather than absolute differences, in this case of wavelengths rather than energies of light.
A second cone system evolved that was most sensitive to the short wavelength region of the visible spectrum, the region we call bluish-violet (Fig. 6). The output of these cones could be compared with the earlier long wave cones that evolved to detect light and dark, and are most sensitive to the yellow-green part of the spectrum (Fig. 6). The spectral sensitivity of a cone is determined by the absorption spectrum of its opsin. The further apart these absorption spectra are, the greater is the potential color contrast (see later Fig. 12). Why not use a red rather than a yellow-green sensitive opsin? There are diminishing returns in using red opsins, probably due to their lower quantal energies. Why not use an ultra-violet opsin? Ultra-violet light is strongly absorbed by the cornea and lens before it reaches the retina and exhibits much chromatic aberration. If these structures are comparatively small as in mice, an ultra-violet cone becomes more tractable.
With two different spectral images of an object and/or its background, the brain can derive differences that are impossible to detect with only one spectral image. Now the brain can distinguish objects, which are not just lighter or darker than their background but have as new attribute, color. An object is uniquely white, gray or black if the two different cone mechanisms are affected equally by the reflected energy of the object and its background. If an inequality exists between the two cone mechanisms, detecting the light reflected by either the object or its background, the brain will see color, making the object or its background appear different than white, gray or black. Objects can have the same brightness as their backgrounds but stand out because of this inequality. This is a powerful way to see objects in a world where most things reflect about the same light as their background.
Nature uses these two cone mechanisms differently. The long wave cones are the sole determinant of light and dark. The short wave cones are only used for color contrast. This strategy minimizes chromatic aberration. For this reason there are about ten times as many long than short wave sensitive cones in most retinas. In the human central fovea there are very few short wave cones and borders are detected by brightness alone.
4. Chromatic versus Achromatic Contrast To distinguish light and dark as well as color, two parallel neural circuits are used in visual cortex. One depends only on long wave cones; it detects differences that depend on light energy. This system is responsible for distinguishing light and dark, i.e. achromatic contrast. The second depends on both long and short wave cones; it compares the response of both cone systems to the same stimulus. This system is responsible for chromatic contrast. Together both systems are responsible for color vision. To distinguish light from dark, neighboring groups of long wave cones are compared for borders of energy contrast; these borders are assembled in the brain as an object (Fig. 7, left). This comparison can involve single neighboring cones or rows of cones at the minimum angle of visual resolution. In the fovea this is a micron, about 1/200th of a degree.
Fig. 7. Neural circuits for Achromatic contrast and Chromatic contrast (59 K jpeg image)
For chromatic contrast, the brain compares the responses of a group of long wave cones with a group of short wave cones in the same retinal area by taking the difference between them (Fig. 7, right). It would be ambiguous to compare only one cone with a neighboring cone because a difference between the two could be caused by energy rather than wavelength gradients. In order to prevent this ambiguity, groups of cones are compared. Such a group of cones can be considered a unit area of chromatic space to distinguish it from a unit area of achromatic space, which in the fovea is a single cone. Therefore achromatic space is more finely divided than chromatic space (Mullen, 1985).
For spatial contrast, the long and short wave cone difference is compared with a difference signal derived from neighboring groups of cones. This creates borders of chromatic (wavelength) contrast. The brain combines borders of chromatic and achromatic contrast to detect objects.
5. Divariant Mammalian Color Vision
About 2% of human males have only two types of cones; long and short wave sensitive. This is a two-variable, i.e. divariant, color vision system similar to many other mammals, such as dogs and cats.
Both the achromatic and chromatic contrast signals are transmitted by long wave cones via the same channels of bipolar and ganglion cells (Fig. 8, left). Small, slowly conducting bipolar and ganglion cells transmit tonic signals of the long wave cones to the brain. One channel is excited by an increment of light absorption by long wave cones; this is an on-channel. Another channel is excited by a decrement of light absorption by long wave cones; this is an off-channel. In the brain both signals are processed by two separate circuits, one of which extracts achromatic and the other chromatic contrast. For achromatic contrast, signals of neighboring tonic long wave cone channels are compared. For chromatic contrast signals from groups of long wave cone channels are compared with those of short wave cone channels.
Fig. 8. Neural circuits for the tonic signals of the long wave cones (59 K jpeg image)
The short wave cones transmit their signals by another tonic system of bipolar and ganglion cells (Fig. 8, right). In this case there are only short wave cone on-channels, presumably because these cones are only involved in chromatic and not in achromatic contrast. In addition these ganglion cells receive and excitatory input from long wave cone off-bipolars. This increases their sensitivity to successive chromatic contrast, i.e. short wave following long wave light.
For chromatic contrast the brain makes two comparisons. It compares the short and long wave cone signals in the same retinal area and then compares this with neighboring areas to detect borders of chromatic contrast.
There is a separate fast, phasic system of larger bipolars and ganglion cells that also transmits signals from long wave cones to the brain (Fig. 9). This system is not as highly developed in the fovea as the tonic system and has a lower spatial resolution. It appears to play no role in color vision. It has a high sensitivity to achromatic contrast (Kaplan and Shapley, 1986) and is sensitive to slow movements (unpublished results). Evidence exists that it contributes to luminance (Lee et al, 1990).
Fig. 9. Neural circuits for the phasic signals of the long wave cones (59 K jpeg image)
6. Simultaneous Contrast
An object's achromatic contrast and brightness depends entirely on the difference between the light it reflects and the light reflected by its surround. Color also depends on simultaneous contrast of wavelength rather than energy contrast. Edwin Land demonstrated the importance of simultaneous contrast in color vision by showing that color depends on the light reflected from the surround of an object. He proposed a model in which the comparison of one cone system with another, needed to arrive at a decision for color, occurs after an object's brightness has been established separately for each cone mechanism. He used the term lightness to describe the brightness of such a monochrome (monocone) image (Land, 1986).
Fig.10. The retinex model of Edwin Land (59 K jpeg image)
Figure 10 illustrates the reasoning behind his retinex (retino-cortex) model. Two projectors cast light on a screen. One transmits white light; the other transmits only long wavelength light. A shadow (arrow) is cast on the screen by blocking the long wave projector (Fig. 10A left). The illumination seen by the two cone systems is shown on the right. The short wave cones see 100% of the white light but 0% of the long wave light. The long wave system sees 100% of the white light and 100% of the long wave light except in the area of the shadow where it sees about 1% of the scattered light. When both projectors are on, the long wave cones detect about 101% of the light from the shadow and 200% from its surround. Therefore the shadow is relatively dark to the long wave cones (Fig. 10A). The retinal area of the shadow seen by the short wave cones absorbs the same light as its surround; therefore it is not dark to these cones. Because the shadow is darker to the long than the short wave cone system, it has a bluish color. This occurs even though more light energy is absorbed from the shadow by the long than the short wave cones.
It is the relative long wavelength gradient of the shadow that determines its color. This decrease in the lightness image of the long wave cones compared to the short wave cones makes the shadow of the arrow, blue (Fig. 10B). Land's algorithm, which computes the lightness of an object for each cone mechanism before comparing these values for color predicts the appearance of colors in a natural scene (Vitek, 1997).
Accordingly, the comparison of the short and long wave cone systems, necessary for color vision, should occur after the normalization for lightness has been established within each cone system. This requires independent processing of the tonic long and short wave cone images before they are compared. In divariant subjects, the long wave cone system is absolutely independent of the short wave cones and therefore satisfies this requirement (Fig.8). The short wave cone system transmits a signal that is relatively independent of the long wave cone system (Fig. 8). It receives input from long wave cone off-bipolars but this only augments its response to short wave light.
How are these separate cone images normalized? Figure 11 provides a reasonable scheme. It is based on so called double-opponent neurons, found in the visual cortex of primates (Michael, 1977). Double-opponent neurons have spatial antagonism between the same cone mechanism mediating the center and surround of their receptive field, something not seen strongly at the level of the retina or geniculate. In this model, long wave cones, responding to the object, excite the neuron and long wave cones, responding to its background, inhibit the neuron. This inhibition normalizes the responses over space because it creates a difference between the responses of different retinal areas, eliminating absolute values (Ratliff, 1960).
After normalization chromatic contrast is achieved by two comparisons. One occurs between neurons reflecting the lightness images of these two cone systems in the same area of space, antagonistically interacting with each other to create neurons responsive to the difference between the two. A second comparison involves this difference signal being compared with those of neighboring areas of visual space to detect simultaneous chromatic contrast.
Figure 11A suggests how a neuron responding to yellow could be formed. It receives an excitatory, normalized input from the long wave cone on- system and an inhibitory input from a normalized short wave on-system. Figure 11B shows a neuron that responds only to blue. It receives an excitatory, normalized input from the short-wave on-system and a normalized, inhibitory input from the long wave cone on-system. The long and the short wave cone systems are compared by subtraction. One difference creates a neuron that is excited by ÒyellowÓ(Fig. 11A). The converse creates a neuron excited by blue (Fig. 11B). These neurons are most sensitive to simultaneous color contrast, the former to yellow on a blue background, the latter to blue on a yellow background. These responses resist changes in the spectral composition of the illuminant, i.e. they show color constancy. They are sensitive to wavelength rather than energy contrast.
White, gray or black is seen when there is no chromatic (wavelength) contrast, i.e. as in a black and white photograph (Fig. 2). The yellow and blue neurons are silent. Another set of neurons is responding. Pure white requires both the long wave and the short wave cone systems to be responding equally. Therefore the long wave on- and the short wave on-systems must both be excited. Black requires both of these systems to be silent and the long wave cone off-system to be excited. There must be separate systems of neurons responding to white (Fig. 11C) and black (Fig. 11D). If both are excited simultaneously, the sensation must be gray.
7. Trivariant Human Color Vision
Normal human color vision depends on three not two cone mechanisms. This adds an additional dimension to color vision creating reds and greens. To do this nature splits the long wave system into two similar systems with slightly different spectral sensitivities with relatively similar opsins (Fig. 12) (Nathans et al., 1986) but the same neural machinery duplicated in parallel circuits. One cone opsin is most sensitive to yellow-green and the other to yellow-red. This splits the brightest and yellow part of the visible spectrum into two color bands, one green and the other red. This red-green system works in parallel with that for blue-yellow.
The splitting of the long wave cone channel into two creates four longer wave channels in the tonic system, two for reddish-yellow and two for greenish-yellow sensitive on- and off-signals (Fig. 13A). The phasic system, which plays no role in color vision mixes both cones in the same on- or off-bipolar cell and retains its original two channels (Figure 13B). In visual cortex, the signals from the tonic reddish-yellow or long wave sensitive (L-) cone and greenish-yellow or middle wave sensitive (M-) cone channels are again used to detect achromatic (energy) and chromatic (wavelength) contrast (Fig. 14A and B).
Trivariance allows the mixing of these parallel red-green and blue-yellow systems to produce non-spectral colors of magenta (red and blue) or cyan (green and blue) (Fig. 15).
Trivariance introduces a second comparison of cone signals, in this case between the two long wave cone systems to create red-green contrast. It occurs in parallel with the divariant comparison of short and long wave cones; in this case both long wave cones are compared to the short wave cones for blue-yellow contrast. If there is a significant difference between the signals of the two long wave cone systems, there will be a hue difference of red or green. If there is no difference between the signals of the two long wave cone systems, the detector will default to the divariant comparison. If there is no difference for the divariant comparison, the signal will appear white, gray or black. It is important to realize that the absorption spectrum of the reddish-yellow cone opsin is not only more sensitive to long wave (red) light but also more sensitive to short wavelength (violet) light than the greenish-yellow cone opsin This gives a reddish color to short wavelengths.
8. Hue, Saturation and Brightness
Color is complex because it depends on more than wavelength contrasts alone (Kaiser & Boynton, 1996). This was appreciated in the 19th century by psychologists who defined the terms hue, saturation and brightness to define color. Hue defines the wavelength contrast aspect of color, such as yellow or blue and red or green. Saturation defines the mixing of hue with white, gray or black. A saturated color has strong hue with little or no white, i.e. blood red. An unsaturated color has its hue washed away by white, i.e. pink. It is reasonable to assume that the independent responsiveness of parallel circuits is responsible for the mixing of hue with white, black or gray. A third quality of color is brightness, which is less obvious than the other two. Yellow and white tend to be bright and blue and black tend to be dark. This difference appears to track the on- and off-systems of the long wave cones. Whenever the on-system is responsive, yellows and whites are likely to be perceived and they tend to be bright. Whenever the off-system is active, blacks and blues are likely to be perceived and they tend to be dark. In addition other qualities of the image such as texture or gloss contribute to color as in gold and silver.
9. The Hering Theory of Color Vision
The Opponent Color Theory of the 19th century physiologist Ewald Hering (Hering, 1964; Hurvich, 1981) derived by the analysis of subjective human color vision is in general correct, although the idea of opponent colors was described earlier by Goethe and Schoepenhauer. Certain colors are not perceived together, i.e. they do not mix. We never see bluish-yellows or reddish-greens. This is consonant with the neural comparisons described previously. The yellow detector is always inactive when the blue detector is active and vice versa. A similar situation occurs for the neurons responding to red or green.
Fig.16. Ewald Hering (59 K jpeg image)
Hering's theory was brilliant but it was proposed when little was known about the anatomy and physiology of the retina. Hering and his school (Ladd-Franklin, 1929) considered that the antagonism between colors occurred in the retina. We now know that color vision is established not in the retina but in visual cortex. The arrangement resembles stereoscopic vision, which is also established in visual cortex, where common signals from each eye are used by different neural circuits to sample objects at different distances in space.
Early recordings of the responses of single neurons in primate retina and geniculate nucleus revealed cells excited by red and inhibited by green light or vice versa (Fig. 17) (Wiesel and Hubel, 1966; Gouras, 1968). These were thought to be the red/green opponent color channel of Hering. Cells were also detected that were excited by blue and inhibited by yellow light or the converse (Fig. 17). These were thought to be the blue/yellow channel of Hering. In addition there were cells which were excited or inhibited by all wavelengths. These were thought to be Hering's white/black channel (DeValois et al., 1966). Now almost half a century later, this view appears to be a misconception. We shall here consider why.
10. Hering Red-Green Channel in the Retina
The cells excited by red and inhibited by green light or the converse, which were considered to be Hering's red-green channel, are comprised of four different types of cells. Two sets are on- and two sets are off-center cells. One set has L-cone on- or off- centers and another has M-cone on- or off-centers. The cells with L-cone centers receive antagonistic signals from M-cones in the surround of their receptive field. The cells with M-cone centers receive antagonistic signals from L-cones in the surround of their receptive field. The strength of this cone-cone antagonism varies considerably.
The first problem with these cells being Hering's red-green opponent channel is that the opponency depends on the geometry of stimulation. Large spots resemble an opponent channel but small spots centered on their receptive field do not, i.e. there is no opponency. The antagonistic surround signal requires more spatial summation than the center mechanism (Fig. 18). A second problem is that a cell with a green sensitive off-center is supposed to be contributing to the sensation of red as much as a cell with a red sensitive on-center. These two cells, which are logically transmitting opposite signals about local brightness, seem inappropriately labeled as a single Hering red-green channel.
A third problem arises when one compares the excitation and inhibition these cells receive from the opposing cone mechanisms. Only a fraction are actually inhibited by red and excited by green light or the converse. Many are excited more by yellow or white light than other colors, which is not really a Hering red-green channel.
There is additional complexity to the L- and M- tonic cone system. It is generally agreed that these cells comprise the midget ganglion cells of the fovea. These retinal ganglion cells receive a direct input from a midget bipolar cell, which receives its input from a single L- or M-cone. The cone antagonism is thought to come indirectly through horizontal and amacrine cells (Fig. 19). There is evidence that the cone antagonism mediated by horizontal cells comes from both L- and M-cones (Dacey, 1996), which is also inconsistent with a Hering red-green opponent channel.
A rival view discounts any role of midget ganglion cells in color vision (Rodieck, 1998; Dacey, 1996). This hypothesis rests on evidence of the existence of a small number of geniculate cells that have coextensive receptive fields with L-cones exciting and M-cones inhibiting or the converse (Wiesel & Hubel, 1966). This result has been difficult to confirm. These cells lack the geometry problem mentioned previously because the red-green opponency does not vary with stimulus size. Coextensive L- and M-cone opponent cells have also been reported to be in the intercalated layers (koniocellular layers) of the geniculate and to project to the areas of visual cortex where double opponent color cells are located (see later figures 25, 26 and 27). These results also need confirmation.
Some support for this hypothesis comes from recordings in the peripheral retina of midget-like ganglion cells, where L- and M- cones are synergistic rather than antagonistic (Dacey and Lee, 1997). This result has been taken to imply that the midget system plays no role in color vision. Another interpretation is that the antagonistic organization of midget ganglion cells is lost outside the fovea because peripheral 'midgets' are not truly midget ganglion cells at all: they receive more than one midget bipolar input and multiple cone inputs (Kolb et al., 1998). So trivariance, may be lost in the peripheral retina.
I believe that all L- and M- tonic cells contribute to color vision but they are not the red-green opponent channels of Hering. I suggest that the antagonism these cells show between L-and M- cone mechanisms is a spectral filter, narrowing the spectral band to which they respond best. This resembles oil droplets located in the cones of certain diurnal vertebrates like birds, reptiles (Fig. 20) and fishes. Oil droplets narrow the action spectrum of the cones, increasing color contrast. Long wave cones become longer wave selective; middle wavelength cones become more middle wavelength selective. This phenomenon makes the monochrome retinex image even more monochrome.
11. Hering Blue-Yellow Channel in the Retina
Some cells excited by blue and inhibited by yellow light or the converse were considered to be Hering's blue-yellow channel. This view is also being challenged nowadays.
The first problem is that most of these cells are excited by light stimulation of short wave cones, either as blue or white light. The Hering blue-yellow channel should not be responsive to white light. The excitation these cells receive from L- and M- off bipolars contributes to successive contrast but exerts little inhibition on the short wave cone signals.
A second problem is their polarity. There are many more excited than inhibited by short wave cones. Malpeli and Schiller (1978) and Gouras and Zrenner (1981) have concluded that retinal and geniculate cells inhibited by short wave cones do not exist. This view is contested by Lee et al. (1987) who have tried carefully to detect cells inhibited by short wave cones. This is not an easy task because cells in the tonic L-and M-cone system can be inhibited by blue and excited by yellow light. In order to eliminate this possibility, Lee et al. (1987) used stimuli, which changed along a tritanopic axis of color space, to which only short wave cones respond. They found cells inhibited by this stimulus. However, macular pigments or other factors could cause their stimuli to miss the tritanopic axis of primate color space thus weakening their conclusion.
Anatomy has revealed that the short wave cone excitatory signal uses a unique bistratified retinal ganglion cell organized to be excited by short wave cones absorbing light and excited by off-responses of longer wave cones (Fig. 21); the converse has not been found. The three dimensional structure of the short wave cone synaptic pedicle reveals a synaptic organization that resembles rod spherules more than longer wave cone pedicles (see chapter on S-cone pathways). The S-cone bipolar dendrites have contacts with S-cones resembling those of on-bipolars of rods. The existence of short wave cone specific off-bipolars is therefore questionable (Kolb et al , 1997).
Electroretinography also suggests that in contrast to the L- and M-cones, there is no evidence of a short wave cone off-bipolar response (Evers & Gouras, 1986).
The short wave on-channel in primate retina is briefly inhibited when a long wave field is turned off, making a blue flash transiently disappear to an observer. This phenomenon was discovered psychophysically by Stiles (1949) and confirmed by Mollon and Polden (1977). This phenomenon is also seen in retinal ganglion cells mediating the excitatory signal of short wave cones (Gouras, 1968), providing a unique psychological marker for the human short wave cone system. A yellow flash does not disappear when a blue adapting field is turned off, which might be expected if there were symmetrical short wave inhibitory-long wave excitory cells. This too supports the absence of a short wave cone off-channel.
Short wave cones have little impact on brightness but have a strong influence on color. If there were short wave cone off-bipolars, they should be logically related to signaling darkness. This would conflict with the L- and M-cones, which signal darkness when their off-bipolar system goes off. In the presence of yellow light the short wave cone off-bipolars would go off when the long wave cone on-bipolars go on. On this reasoning short wave cone off- channels are inappropriate.
12. Hering White-Black Channel in the Retina
Although these sensations are opposite in nature, they are not opponent as blue and yellow or red and green colors are. The intermediary sensation of gray is a mixture of black and white. The reason for this is that no antagonism occurs between cone mechanisms for the establishment of white and black. These sensations depend on antagonism in space but not among cones. It is not a wavelength but an energy comparison. White or gray depends on all cone mechanisms absorbing light and for pure white or gray this absorption rate must be relatively equal. If the short wave channel is not active, the whites and grays become yellowish, greenish or reddish.
The early recordings of single retinal and geniculate neurons in monkeys revealed a subset of cells that were either excited or inhibited by all wavelengths. These were considered to represent the white and black channels of Hering. At that time, there was no distinction between tonic (parvo) and phasic (magno) systems. The idea that these cells reflect the perception of white and black is incorrect because they respond to all colors, not just white or black. The perception of white and black occurs in the visual cortex, where simultaneous contrast occurs. I suggest that pure white occurs when the normalized images of the three cone on-systems are equal for the object. Pure black occurs when the normalized L-and M-cone off -systems are equally excited and the short wave cone on-system is silent in the representation of the object in the cortex.
13. Retinal Interneurons
The horizontal and amacrine cells represent laterally interacting elements in the retina. In general they are inhibitory (antagonistic) in their neural interactions. There are exceptions, notably the rod amacrine cell, which transmits rod signals to bipolars and ganglion cells (see chapter on circuitry for rod signals).
Among primate horizontal cells there are at least two major classes. One class receives its input overwhelmingly from L- and M-cones. A second class receives its input from all three cones but much more from short wave ones (Fig. 22) (Ahnelt and Kolb, 1994; Dacey, 1996). It is thought that horizontal cells feedback antagonistically on cones. The specificity of the feedback is not clearly defined. It is possible that the horizontal cells receiving inputs from short wave cones only exert feedback on to short wave cones, even though they receive inputs from L- and M-cones. I favor this hypothesis because one does not see any input from short wave cones in ganglion cells receiving inputs from L- and M-cones. If the horizontal cells receiving inputs from short wave cones were to feedback on to L- and M-cones, short wave cone inputs would be found in all retinal ganglion cells, which is not the case.
What is the role of horizontal cells? It would seem that the negative feedback they exert on the cones is to control overdriving of the cones by large energy gradients in both space and time. Over-stimulation of the cones would be curtailed and response speed increased. Spatial contrast could also be facilitated (Ratliff, 1960). The horizontal cell feedback is only brought into action by strong stimuli that affect large groups of neighboring cones. This could also enhance spectral contrast. The depolarization that yellow light would exert on S-cones through horizontal cells could enhance any subsequent hyperpolarization produced by short wave light. This would augment responses to short wave light on long wave backgrounds and facilitate successive color contrast.
The amacrine cells are less understood than horizontal cells. It is reasonable to assume that most transmit antagonistic interactions to bipolar and ganglion cells. The amacrine cell interaction occurs after the on- and off- systems of bipolar cells are established. This allows separate channels of antagonism to be mediated by on- and off- amacrine cells. Their role would be similar to horizontal cells but at the inner plexiform layer (Fig. 23).
Another role of the amacrine cell system is to help establish the functional differences in the tonic and phasic ganglion cell systems. The complete circuitry of these two systems is not known. It is likely that the phasic system has separate on- and off- bipolars that transmit signals from cones to the ganglion cell layer (Fig.13) because midget cone bipolars synapse only on midget ganglion cells and not on other ganglion cells (Kolb, 1994; Boycott and Wassle,1999). Therefore, another system of L- and M-cone bipolars, presumably diffuse cone bipolar types, (see chapter on cone pathways through the retina) must transmit cone signals to phasic ganglion cells. It would be at these bipolars and/or ganglion cells that amacrine cells establish the phasicity of the phasic ganglion cells (Werblin, 1991; Slaughter et al., 1995; Cook and McReynolds, 1998). L- and M-cone signals reach these ganglion cells faster than they reach the tonic ganglion cells (Gouras, 1968), which also suggests a separate bipolar system.
14. The Role of Phasic Ganglion Cells
The first recordings from cells in the magno-cellular layers of the lateral geniculate of monkeys were by Wiesel and Hubel (1966). The sample of cells they encountered in these layers was relatively small. They described cells that were tonically inhibited by red light, a phenomenon not found in their more extensive sampling of the parvo-cellular layers.
Two different classes of ganglion cells occur next to each other in monkey retina, one phasic, the other tonic (Gouras, 1968; DeMonasterio and Gouras, 1975). These differences are striking in the retina where these two different types of cells can be recorded from simultaneously. Antidromic driving showed that phasic cells have faster conduction velocities than tonic ones and therefore were presumably larger cells, suggesting a link with the magno- and parvo-cellular layers of the geniculate, respectively. The antidromic field potential of these two cell groups revealed that the tonic cell system was concentrated around the fovea while the phasic system was evident perifoveally and peripherally (Gouras, 1969).
It is difficult to know what visual sensation the phasic system mediates. It has been suggested that it is responsible for luminance (Lee et al., 1990). Luminance is an additive quality, which has an action spectrum reflecting the combined L-and M- cone systems but not short wave cones. It has also been suggested that the phasic system detects movement. One of the reasons for this hypothesis is that the phasic ganglion cell system projects to the magno-cellular layers of the lateral geniculate nucleus, which in turn projects through striate cortex to a visual area MT, where cells sensitive to movement and the direction of movement seem to be found. This idea is not completely accepted. It is also possible that the phasic system mediates the signal for the optokinetic reflex, responding to very slow retinal movements (unpublished observations).
The phasic system is not involved in color vision because it synergistically mixes the signals of L- and M-cones, which is not what one expects from a system involved in color vision. It is possible that the signal of the phasic system is used for blue-yellow contrast. However, I doubt this hypothesis because 1) the retinal distribution of the phasic ganglion cells is not in register with that of the tonic system and 2) their latencies and conduction velocities are out of phase with those of the tonic system.
Why should one visual pathway be phasic and the other tonic? This is an intriguing question that awaits more research. It has been suggested that the foveally oriented tonic pathway plays a more dominant role in conscious perception in which tonic responses may be an advantage (Martinez-Conde et al., 1999).
15. The Lateral Geniculate Nucleus
The lateral geniculate nucleus (see Fig. 3) forms the main stream of visual information to the cerebral cortex. It is a transfer center that disentangles the various retinal subsystems serving the contralateral visual fields and organizes their projections to striate cortex.
The tonic system, carrying cone specific channels for both high visual resolution and color vision, is confined to the parvo-cellular layers (Fig. 25). The phasic system is confined to the magno-cellular layers (Fig. 25). The signals from each eye are kept separate in order to combine them appropriately for stereoscopic vision, where different combinations reflect different depth planes. Similarly on-cells are separated from off-cells and cone specific responses are kept separate for color vision. There are non-retinal inputs to the lateral geniculate nucleus that must modulate the flow of information but they are poorly understood. They may play a role in attention and sleep.
16. Color Vision in Visual Cortex
The cerebral cortex mediates conscious perception. In striate cortex there are local zones called blobs (Fig. 27) that contain cells that exhibit double-opponent behavior (Michael, 1977; Livingstone and Hubel, 1988). We assume double-opponency is an essential stage in color vision. Double-opponency is not introduced at an earlier stage because a commitment to color contrast cannot be made until the achromatic information is extracted from the tonic L- and M-cone input channels.
Achromatic and chromatic information is considered to be multiplexed in the tonic L- and M-cone system, which is demultiplexed before chromatic contrast is established. By demultiplexing is meant that chromatic information is transmitted to different neural circuits than those for achromatic information (Fig. 27). The L-and M-cone on- and off-channels are used synergistically for achromatic contrast detectors. These same channels are used antagonistically for chromatic contrast detectors. Achromatic contrast involves larger populations of neurons located outside the smaller blob areas (Fig. 27, achromatic and oriented).
There is no multiplexing for the short wave channel because it is totally committed to color. Thus this channel goes directly from LGN to the blobs (Fig. 27). Double-opponency involving the short wave cone system also depends on the demultiplexing of the longer wave cone system's signal before a comparison is made for chromatic contrast.
We assume that chromatic contrast begins with double-opponent cells (Fig. 11). Double-opponency establishes wavelength contrast independently of brightness contrast. These two forms of contrast become independent neural entities in establishing borders of contrast.
Orientation selectivity, undoubtedly essential for form vision, could be based on either or both of these forms of contrast. The greater spatial resolution of achromatic contrast appears to involve more orientation selective cells in visual cortex than color contrast does.
17. Color Vision beyond Striate Cortex
Knowledge of the physiology of color vision beyond striate cortex is sketchy because of the difficulty in exciting cortical cells in anesthetized monkeys. Post striate visual cortex is composed of a number of separate areas. One area, called MT, appears to receive its major input from the magno-system. Several other areas, called visual areas 2, 3 and 4 receive their input from the parvo-system. The role of these different areas is unclear (Schiller and Lee, 1991). Color selective cells have been found in areas 1, 2, 3 and 4 (Kruger and Gouras, 1979), although Zeki (1993) has maintained that they are much more numerous in visual area 4 (V4).
Using three projectors to control the spectral reflectance of objects in a Mondrian-like display, Zeki (1993) reported a major difference between striate cortex and visual area 4. Spectrally selective cells in striate cortex of an anesthetized monkey responded to the energies reflected from the objects and not to their color. In other words, objects with different colors but reflecting identical energies from their surface evoked the same response when presented to such neurons. On the other hand, spectrally selective cells in area V4 responded to the colors and not to the energies. This implies that color is not established in striate cortex but at a later stage in vision.
Zeki has raised the interesting hypothesis that V4 is the visual center for color and is not represented in a retinotopic but in a chromatic coordinate system. An object's shape would be defined in area V1 (striate cortex) and perhaps V2 where the anatomy reflects a retinotopic plan, and its color would be defined in V4 following a chromatic plan. In V4 colors would be distributed in an orderly way over the structure somewhat like a chromaticity diagram (Fig. 27). This is a radical view of cortical processing that needs more research. A problem with this hypothesis is how a neural response that represents an object in one cortical area can be linked with its color response in another area. Color is complex depending not only on hue, defined by the chromaticity diagram but also by achromatic cues such as shadowing and texture. Another view is that chromatic contrast detectors are ubiquitous in all visual areas, providing parallel cues to borders with achromatic contrast detectors. The remapping of visual space, occurring in these multiple visual areas may increase the universe of forms and objects distinguishable (Fig. 28).
The Commision Internationale d'Elairage (CIE) chromaticity diagram (Figure 27) defines human trivariant color experience in a quantitative way, which allows standards for color comparisons worldwide. It is defined by three spectral variables, which can be combined to mimic any color. The values of the three standards have been normalized so that they always add up to 1.0. In this way the color diagram can be plotted in two dimensions, those of the long (red) and middle (green) wave standards. What does not equal 1.0 on the diagram represents the short wave standard. In the middle of the diagram, where the three standards are all 0.33, the color is white (gray or black). Moving up the diagram makes the color green, right makes the color red and down makes it blue. The value 1.0 can never be reached because the absorption spectra of the three cone opsins overlap so much that the standards stimulate more than one cone mechanism. Therefore, all colors are to some extent desaturated. Perhaps if one could find a way to stimulate only one class of cones, more saturated colors might be perceived. Normal humans can perceive about a million different colors but this number depends on the right conditions. The illumination must be bright and the size of the color stimulus must be large, at least 20 degrees of visual angle to appreciate such a large universe of colors. Why very large stimuli create a much greater palette of colors must depend on spatial summation in visual cortex (Figure 28).
18. Color and Form
One of the most impressive facts about the physiology of visual cortex is the relatively small proportion of cells that have any striking wavelength selectivity. This is in marked contrast to the retino-geniculate pathway where many cells show wavelength selectivity. This suggests that much of the neural machinery in visual cortex is used for pattern recognition rather than color vision, and that wavelength selectivity plays only a minor role in the response repertoire of cortical visual neurons. Evidently the processing of a black and white (achromatic) image requires almost as much neural power as the processing of a color image: perhaps like a color television, where much more information is transmitted for the pattern than for the color of the image.
Just where color enters into pattern recognition will require a better understanding of the latter. It would seem that objects are identified by connected groups of orientation selective neurons having appropriate contrasts (Marr, 1982) and these objects are reinforced by their structural stability with movement in space. In this, light, dark and shading have considerable impact on contour stability and structure as evidenced by the pattern recognition possible in black and white photographs. Color adds an important but minor frill to this complex neural construct but because of its limited dimensionality (only two or three dimensions), it has become an alluring femme fatale to many of us.
19. References.
Ahnelt, P. and Kolb, H. (1994) Horizontal cells and cone photoreceptors in human retina: A Golgi-electron microscopic study of spectral connectivity. J. Comp. Neurol. 343, 406-427.
Bowmaker, J. K. (1998) Evolution of color vision in vertebrates. Eye 12,541-547.
Boycott, B. B. and Wassle, H. (1999) Parallel processing in the mammalian retina. Invest. Ophthal. Vis. Sci. 40,113-1327.
Cook, P. B., and McReynolds, J. S. (1998) Lateral inhibition in the inner retina is important for spatial tuning of ganglion cells. Nature Neuroscience 1, 714-9.
Dacey, D. M. (1996) Circuitry for color coding in the primate retina. Proc. Nat. Acad. Sci. 93, 582-588.
Dacey, D. M. and Lee, B. B. (1997) Cone inputs to the receptive fields ofmidget ganglion cells in the periphery of Macaque retina. Invest. Ophthal. Vis. Sci 38, S708.
Dacey, D. M. and Lee, B. B. (1994) The blue-on opponent pathways in primate retina originates from a distinct bistratified
ganglion cell. Nature 367, 731-735.
DeMonasterio F. M. and Gouras, P. (1975) Functional properties of ganglion cells of the rhesus monkey retina. J. Physiol. (Lond) 251,167-95.
DeValois, R. L., Abramov, I. and Jacobs, G. H. (1966) Analysis of response patterns of LGN cells. J. Opt. Soc. Am. 56, 966-977.
Dowling, J. E. (1987) The Retina: an approachable part of the brain. The Belknap Press, Harvard University Press Cambridge, Massachusetts.
Evers, H. and Gouras, P. (1986) Three cone mechanisms in the primate electroretinogram: two with and one without off-center bipolar responses. Vision Res. 26, 245-254.
Gouras, P. (1968) Identification of cone mechanisms in monkey ganglion cells. J. Physiol. 199, 533-547.
Gouras, P. (1969) Antidromic responses of orthodromically identified ganglion cells in monkey retina. J. Physiol. 204, 407-419.
Gouras, P. and Zrenner E. (1981) Color vision: a review from a neurophysiological perspective. Prog. in Sens. Physiol. 1,139-179.
Hering, E. (1964) Outlines of a theory of the light sense. Translated by L.M. Hurvich & D. Jameson. Cambridge, MA: Harvard University Press.
Hurvich, L.M. (1981) Color Vision. Sunderland, MA: Sinauer Assoc.Inc.
Kaiser, P. and Boynton, R. M. (1996) Human Color Vision. Opt. Soc. of Amer.
Kaplan, E. and Shapley, R. M. (1986) The primate retina contains two groups of ganglion cells, with high and low contrast sensitivity. Proc. Natn. Acad. Sci. USA 83, 2755-2757.
Kolb, H. (1994) The architecture of functional neural circuits in the vertebrate retina. Invest. Ophthal. Vis. Sci. 35, 2385-2404.
Kolb, H., Goede, P., Roberts, S., McDermott, R., and Gouras, P. (1997) Uniqueness of the S-cone pedicle in the human retina and consequences for color processing. J. Comp. Neurol. 386, 443-460.
Kolb, H., DeKorver, L., Church, J., Crooks, J., Jacoby, R. and Marshak, D. (1998) P cells of the primate retina. Invest. Ophthal. Vis. Sci. 39, S563.
Kruger, J. and Gouras, P. (1979) Spectral selectivity of cells and its dependence on slit length in monkey visual cortex. J. Neurophysiol. 43, 1055-1069.
Ladd-Franklin (1929)
Land, E. (1986) Recent advances in retinex theory. Vision Res. 26.7-21.
Lee, B. B., Pokorny, J., Smith, V. C., Martin, P. R., and Valberg, A. (1990) Luminance and chromatic modulation sensitivity of macaque ganglion cells and human observers. J. Opt. Soc. Am. 7, 2223-2236.
Lee, B. B., Valberg, A., Tigwell, D. A. and Tryti, J. (1987) An account of spectrally opponent neurons in macque lateral geniculate nucleus to successive contrast. Proc. Roy. Soc. (Lond.) B 230, 293-314.
Livingstone, M. S. and Hubel, D. H. (1988) Segregation of form, color, movement and depth: Anatomy, physiology and perception. Science 240, 740-749.
Malpeli, J. G. and Schiller, P. (1978) Lack of blue off-center cells in the visual system of the monkey. Brain Res. 141. 385-389.
Marr, D. (1982) Vision. San Francisco W. H. Freeman & Co.
Martinez-Conde, S., Macknik, S. L. and Hubel, D. H. (1999) Bursts of spikes in the LGN and area V-1 are correlated with microsaccades during visual fixation in the behaving monkey. Invest. Ophthalmol. Vis. Sci. 40, S642.
Michael, C. (1977) Color vision mechanisms in monkey striate cortex: dual opponent cells with concentric receptive fields. J. Neurophysiol. 41, 572-588, 1987.
Mollon, J. D. and Polden, P. G. (1977) An anomaly in the response of the eye to light of short wavelength. Phil. Trans. Roy. Soc. (Lond.) B 238 207-240.
Mullen, K. T. (1985) The contrast sensitivity of human colour vision to red-green and blue-yellow chromatic gratings. J. Physiol. 359, 381-400.
Nathans, J., Thomas, D., and Hogness, D. S. (1986) Molecular genetics of human color vision: The genes encoding blue, green and red pigments. Science 232, 203-210.
Ratliff, F. (1960) Mach Bands: Quantitative studies on neural networks in the retina. Holden Day Inc. San Francisco.
Rodieck, R. W. (1998) The First Steps in Seeing. Sunderland, MA: Sinauer Assoc.
Schiller, P. H. and Lee, K. (1991) The role of primate area V4 in vision. Science 251,1251-1253.
Schiller, P. H., Logothetis, N. K. and Charles, E. R. (1990) Functions of the the color-opponent and the broad-band channels in vision. Visual Neurosci. 5, 321-346.
Schiller, P. H., Logothetis, N. K. and Charles, E. R. (1991) Parallel pathways in the visual system: their role in perception. Neuropsychologia 29, 433-442.
Slaughter, M.M, Zhang, J. & Tian, N. (1995) Ramifications of GABA receptor subtypes on retinal information processing. In J. Robbins (ed): Basic and Cilinical Perspectives in Vision Research. New York: Plenum Press, pp. 115-124.
Stiles, W.S. (1949) Increment thresholds and the mechanisms of colour vision. Documenta Ophth. 3,138-165.
Vitek, (1997)
Walls, G. L. (1942) The vertebrate retina and its adaptive radiation. Cranbrook Press, Michigan.
Werblin F. S. (1991) Synaptic connections, receptive fields, and patterns of activity in the tiger salamander retina. Invest. Ophthal. Vis. Sci., 32, 459-483.
Wiesel, T. and Hubel, D. H. (1966) Spatial and chromatic interactions in the lateral geniculate body of the rhesus monkey. J. Neurophysiol. 29,1115-1156.
Zeki, S. A (1993) Vision of the Brain. Oxford, Blackwell Scientific.
|
The author Peter Gouras was born in Brooklyn, New York. He received an MD degree from Johns Hopkins University School of Medicine and proceeded directly into a career in eye research at the National Institutes of Health, as part of the Public Health Service, at first in the Neurological Institute and the Division of Blindness and later in the National Eye Institute. Dr. Gouras is renowned in 3 areas of vision research all of which he has been a pioneer and leader. For most of his career he has been active in the fields of basic and clinical research on the ERG. In the late sixties he pioneered recordings of primate ganglion cells in the fovea and did pivotal work on the analysis of color pathways from retina through LGN to visual cortex. On his moving to the faculty of Harkness Eye Institute at Columbia Medical School, New York City, in the mid 1970s, Dr Gouras made daring strides into the area of pigment epithelium transplantation for the treatment of retinal and macular disease. He is the author of several hundred papers and many chapters and books particularly concerning color vision and retinal information processing. |
 |
[Introduction] [The Evolution of Vision] [Color Vision] [Chromatic versus Achromatic Contrast] [Divariant Mammalian Color Vision] [Simultaneous Contrast] [Trivariant Human Color Vision] [Hue, Saturation and Brightness] [The Hering Theory of Color Vision] [Hering Red-Green Channel in the Retina] [Hering Blue-Yellow Channel in the Retina] [Hering White-Black Channel in the Retina] [Retinal Interneurons] [The Role of Phasic Ganglion Cells] [The Lateral Geniculate Nucleus] [Color Vision in Visual Corte] [Color Vision beyond Striate Cortex] [Color and Form] [References]
January, 2001
