Open Access
Add to BibSonomy
Abstract
The irreducible unique hues—red, green, blue, and yellow—remain one of the great mysteries of vision science. Attempts to create a physiologically parsimonious model that can predict the spectral locations of the unique hues all rely on at least one post hoc adjustment to produce appropriate loci for unique green and unique red, and struggle to explain the non-linearity of the Blue/Yellow system. We propose a neurobiological color vision model that overcomes these challenges by using physiological cone ratios, cone-opponent normalization to equal-energy white, and a simple adaptation mechanism to produce color-opponent mechanisms that accurately predict the spectral locations and variability of the unique hues.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. INTRODUCTION
Human color vision is encoded by long (L), middle (M), and short (S) wavelength-sensitive cones in the retina [1]. However, when modulated in isolation, the ${\rm{L}}$, ${\rm{M}}$, and ${\rm{S}}$ cones do not produce the pure color sensations of red, green, and blue; ${\rm{L}}$-cone isolating stimuli appear yellowish-red (salmon), ${\rm{M}}$-cone isolating stimuli appear bluish green (teal), and S-cone isolating stimuli appear reddish blue (violet).
Perceptually, humans experience not three, but four phenomenologically unmixed, or “pure,” color sensations—red, green, blue, and yellow—which are explained by the activity of two opposing dichromatic systems: Red versus Green (R/G) and Blue versus Yellow (B/Y) [2,3]. Consequently, colors that require combining both members of one dichromatic system are impossible—there is no reddish green or bluish yellow—and whenever the two opponent sides of one system are in equilibrium and the other is polarized in one direction or another, we experience the pure sensations referred to as the “unique hues.” The wavelength at which the B/Y system crosses zero is unique green, while the zero-crossings of the R/G system are the wavelengths of unique blue and unique yellow. Experimentally, the wavelengths of these unique hues are measured for isolated spots of light against a dark background and not in the more general domain of color appearance in natural settings.
Measurements from the primate lateral geniculate nucleus (LGN) by DeValois, Abramov, and Jacobs in 1966 established the existence of cone-opponent mechanisms that compare ${\rm{L}}$ versus ${\rm{M}}$ cones and ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) cones [4], but these cone-opponent mechanisms do not line up with the color-opponent mechanisms originally proposed by Hering [2], whose zero-crossings define the unique hues. Therefore, how the visual system transforms the spectral sensitivities of the ${\rm{L}}$, ${\rm{M}}$, and ${\rm{S}}$ cones into perceptual R/G and B/Y color-opponent mechanisms with the appropriate spectral positions and chromaticity coordinates for the unique hues remains one of the great unsolved mysteries of vision science [5].
What is required for the equilibrium hues to be modeled as a linear system was developed with mathematical rigor by Krantz (1975) [6], and linear models have been tested empirically by Kranz, Larimer, and Cicerone (1974–1975) [7–9]; however, beyond the cones, such models did not explicitly specify the neurobiological underpinnings. Several multi-stage models that attempt to reconcile the ${\rm{L}}$ versus ${\rm{M}}$ and ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) cone-opponent mechanisms with the perceptual color-opponent mechanisms [10–12] have been proposed, and DeValois and DeValois (1993), in particular, propose detailed neurobiological underpinnings for the first two stages [10], but, as Conway and colleagues recently pointed out, a simple linear combination of the former has never successfully produced the later [13]. All existing models require at least one post hoc adjustment to produce a B/Y system that accurately predicts the spectral location of unique green, and no truly linear system can account for the fact that unique red and unique green are not co-linear in color space, indicating that the B/Y system is inherently nonlinear [14–17].
In this paper, we propose a neurobiological three-stage color vision model to explain the neural basis of the unique hues without relying on any post hoc adjustments to fit the psychophysical data. By taking into account the wide variability in L:M cone ratios and normalizing second-stage cone-opponent mechanisms to equal-energy white, the model generates accurate color-opponent mechanisms that predict both the position and variability of the spectral unique hues—blue, green, and yellow [14,18]—without any further adjustments. Furthermore, by taking differential adaptation of the ${\rm{S}}$ and ${\rm{L}}$ cones at short and long wavelengths into account, we can explain the non-linearities in the R/G and B/Y color-opponent mechanisms demonstrated by Burns et al. [16].
In the model presented here, the ${\rm{L}}$, ${\rm{M}}$, and ${\rm{S}}$ cones constitute the first stage, the spectral sensitivities of which have been adjusted to account for absorption by the lens and macular pigment. The second stage combines the spectral sensitivities into ${\rm{L}}$ versus ${\rm{M}}$ and ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) cone-opponent mechanisms using the relative numerosity of the ${\rm{L}}$ and ${\rm{M}}$ cones in the retina to determine the spectral sensitivity of the (${\rm{L}} + {\rm{M}}$) side of the ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) mechanism. These second-stage mechanisms are normalized to equal-energy white. Finally, the cone-opponent mechanisms are linearly combined at the third stage to produce two color-opponent mechanisms—(${\rm{S}} + {\rm{M}}$) versus ${\rm{L}}$ for Blue versus Yellow and (${\rm{S}} + {\rm{L}}$) versus ${\rm{M}}$ for Red versus Green—that accurately predict the spectral positions of the unique hues.
By adjusting the L:M ratio or macular pigment density, our model also captures the relative variability of the unique hues: predicting the extreme variability of unique green and the relative stability of unique blue and unique yellow among color normal observers. Finally, the model also predicts the spectral neutral points of Protanopes and Deuteranopes [19,20], which lie at the cross-points of the normalized second-stage ${\rm{S}}$ versus ${\rm{M}}$ and ${\rm{S}}$ versus ${\rm{L}}$ opponent mechanisms.
2. METHODS
A. Cone Spectral Sensitivities
The spectral sensitivities of the cones were derived using a physiologically based photopigment template [21], with the ${\rm{S}}$, ${\rm{M}}$, and ${\rm{L}}$ cone photopigments defined by spectral peaks at 419, 530, and 557.25 nm and optical densities of 0.4, 0.22, and 0.34, respectively, reflecting the average spectral peaks for the ${\rm{S}}$, ${\rm{M}}$, and ${\rm{L}}$ cones based on molecular genetics and flicker photometric electroretinogram (ERG) [22,23].
To account for the rapid decline in macular pigment density with eccentricity from the fovea, we estimated the relationship between eccentricity and macular pigment density using the measurements of Hammond and colleagues [24]. The model allows us to adjust the amount of macular pigment filtering either manually, as a percentage, or by estimating the percent-of-max at a given retinal eccentricity, as shown in Fig. 1(a). Thus, adjustments to macular pigment filtering produce different cone spectral sensitivities depending on the size (in degrees) of our hypothetical stimulus, which is a disc of light centered on the fovea presented against a black background.
Fig. 1. (a) Average macular pigment absorption as a function of eccentricity based on measurements by Hammond et al. [24]. (b) ${\rm{S}}$, ${\rm{M}}$, and ${\rm{L}}$ cone spectral sensitivities derived using a physiologically based photopigment template [21] and corrected for lens and macular pigment absorption [25] corresponding to a 2 degree stimulus.
After correcting for optical filtering using the Stockman lens [25], and accounting for the proportion of macular pigment filtering for a 2 degree stimulus, putative spectral sensitivities for the cones, as measured at the cornea, are produced as shown in Fig. 1(b).
B. Cone-Opponent Mechanisms
Once appropriate cone spectral sensitivities have been produced, ${\rm{L}} - {\rm{M}}$, ${\rm{M}} - {\rm{L}}$, and ${\rm{S}} - ({\rm{L}} + {\rm{M}})$ cone-opponent mechanisms are derived and normalized so that they null to equal-energy white. This was done using the lsqnonlin function in MatLab (version R2022b) to derive the appropriate von Kries coefficients for each component [26], with the (${\rm{L}} + {\rm{M}}$) surround of each mechanism treated as a single entity, using the L:M cone ratio as individual ${\rm{L}}$ and ${\rm{M}}$ cone weights.
Assuming an L:M ratio of 2:1, the second-stage mechanisms are derived such that
This is a hybrid between the model proposed by DeValois and DeValois (1993) [10], which relies heavily on cone ratios for its calculations, and the Stockman and Brainard model [12], which relies on normalization to equal-energy white in order to achieve appropriate zero-crossings for the unique hues. Nulling to environmental white ensures that the color-opponent system is silent in response to a white stimulus [27]; however, the Stockman and Brainard model assumes this normalization happens at the third stage, when cone-opponent signals are recombined into color-opponent mechanisms [12]. Applying the normalization at the second stage and then combining the normalized cone-opponent mechanisms as described below is one of the choices that sets our model apart by allowing this important normalization to happen at the level of horizontal feedback at the cone pedicle.
Using these parameters, we produce the normalized cone-opponent mechanisms shown in Fig. 2.
Fig. 2. ${\rm{L}}$ versus ${\rm{M}}$ (red), ${\rm{M}}$ versus ${\rm{L}}$ (green), and ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) (blue) cone-opponent mechanisms produced by combining the cone spectral sensitivities in Fig. 1(b) and normalizing to equal-energy white using Eqs. (1 –3).
C. Color-Opponent Mechanisms
The third-stage R/G and B/Y color-opponent mechanisms are produced by subtracting the normalized cone-opponent mechanisms in Fig. 2 from one another: ${\rm{L}}$ versus ${\rm{M}}$ is subtracted from ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) to produce the B/Y mechanism while ${\rm{M}}$ versus ${\rm{L}}$ is subtracted from ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) to produce the R/G mechanism. Because each cone-opponent mechanism is already normalized to white, no further adjustments are required. Both color-opponent mechanisms already have a null response to equal-energy white, as shown in Fig. 3, and both the R/G and B/Y mechanisms now contain inputs from all three cone types.
Fig. 3. Red versus Green (red) and Blue versus Yellow (blue) color-opponent mechanisms produced by subtracting the normalized cone-opponent mechanisms in Fig. 2 from one another.
The zero-crossings of our R/G system represent the spectral locations of unique blue and unique yellow. The zero-crossing of our B/Y system represents the spectral location of unique green. For desaturated colors of constant hue, the proportion of each primary required to null either the R/G or B/Y system can be represented on the Judd (1951a) chromaticity diagram by convolving these proportions with the corresponding tristimulus values [25].
The shape and zero-crossings of the color-opponent mechanisms produced by our model are dependent on three factors: the gains of the second-stage cone-opponent mechanisms, which are set by long-term visual experience (i.e., normalized to equal-energy white), the ratio of ${\rm{L}}$ to ${\rm{M}}$ cones in the retina, and the density of the macular pigment. All these variables can be adjusted independently in our MatLab live code (available at [28]) in addition to the spectral peaks and optical densities of the input cones and environmental white used to normalize the cone-opponent signals. By default, and wherever we reference “average” unique hues in this paper, the L:M ratio is set to 2:1, the stimulus size for macular pigment filtering is set to 2° (corresponding to approximately 50% of maximum), the cones are characterized by the spectral peaks and optical densities described in Section 2.A, and the normalizing environment is set to equal-energy white.
D. Adaptation
It is well established that unique green and unique red are not co-linear in color space [14–17], rendering any truly linear model of color-opponency inaccurate because it can’t account for the non-linearity present in the Blue/Yellow mechanism and, to a lesser extent, the Red/Green mechanism [16].
To account for this non-linearity, we have implemented an adaptation paradigm that differentially adjusts the ${\rm{L}}$ and ${\rm{S}}$ cone spectral sensitivities in proportion to their quantal catches relative to the ${\rm{M}}$ cone for any given mixture of primary lights. This is possible because our third-stage color-opponent mechanisms are made up of inputs from all three cones and differentially adapting the ${\rm{L}}$- or ${\rm{S}}$-cones with long- or short-wavelength light, respectively, will shift the color opponent curves and corresponding unique hue loci. For a given spectral primary, the cones are differentially adapted according to Weber’s Law [29],
This approach has almost no effect across most of the chromaticity space, where the quantal catch of the ${\rm{M}}$ and ${\rm{L}}$ cones is relatively close to 1:1 and the ${\rm{S}}$ cone is nearly silent. However, the effects become significant at long and short wavelengths when the ${\rm{L}}$ and ${\rm{S}}$ cones, respectively, are being driven much more strongly, decreasing their sensitivity relative to the other cones. These changes, which happen on a very short time scale, can affect the B/Y and R/G color-opponent mechanisms significantly.
In Fig. 4, the unadapted B/Y mechanism (solid line) is plotted alongside the adapted mechanism produced by proportional adaptation to 45% 470 nm and 55% 610 nm light (dotted line) and 25% 470 nm and 75% 640 nm light (dashed line). These are the proportions required to null the unadapted B/Y color-opponent mechanism, but our model predicts that the long-wavelength primary will have a significant impact on the sensitivity of the ${\rm{L}}$ cone.
Fig. 4. Blue/Yellow color-opponent mechanism with no adaptation (solid lines), adapted using a combination of 45% 470 and 55% 610 nm light (dotted line), and adapted using a combination of 25% 470 and 75% 640 nm light (dashed line).
To plot chromaticity coordinates for desaturated colors of constant hue, we calculated the ${\rm{L}}$- and ${\rm{S}}$-cone adaptations for a given pair of spectral primaries as described previously, recalculated the null using the adapted color-opponent curves, and convolved the resulting proportions with the Judd (1951a) tristimulus values. This chromaticity space was chosen to facilitate direct comparison with the psychophysical data from Burns et al. [16].
3. RESULTS
A. Spectral Position and Variability of Unique Hues
Using an L:M ratio of 2:1, a 2 degree stimulus centered on the fovea, and equal energy as our environmental white, the R/G and B/Y color-opponent mechanisms produced by our model predict average positions for the three spectral unique hues: blue, green, and yellow. The zero crossings of the Red versus Green system put unique blue and unique yellow at 476.7 and 579.2 nm, respectively, while the zero crossing of the Blue versus Yellow system puts unique green at 518.3 nm. These results agree well with real-world measurements of the unique hues gathered by Schefrin and Werner [18] and Jordan and Mollon [30,31].
The model also predicts that the observed difference in the relative variability of these hues will be a function of the variability in L:M ratio and macular pigment filtering. Unique green can vary by over 50 nm among color normal observers [31–34], while unique blue and unique yellow are relatively stable. Changing our model’s L:M ratio and/or the macular pigment density produces large shifts in unique green but only modest changes in unique blue and unique yellow.
If we adjust the L:M ratio from 1:1 to 10:1, unique green will shift from a maximum of 534.8 nm down to a minimum of just 499.5 nm, while unique blue and unique yellow shift by only 4.7 and 5.6 nm, respectively, as shown in Fig. 5.
Fig. 5. Color-opponent mechanisms produced by our model with the L:M ratio set to 1:1 (solid lines) and 10:1 (dotted lines), with macular pigment filtering corresponding to a 2 degree stimulus (see Fig. 1).
A similar shift is produced when adjusting macular pigment density. According to measurements made by Hammond and colleagues, macular pigment density falls off very sharply with eccentricity [24], dropping to 50% of maximum by 1° and 10% of maximum by 2°, on average. In our model, decreased macular pigment filtering pulls unique green to shorter wavelengths. If we keep L:M ratio constant at 2:1 and adjust the macular pigment density from 100% down to 10%, unique green will shift from 535.4 to 507.4 nm, while unique blue and unique yellow shift by just 0.1 and 3.3 nm, respectively, as seen in Fig. 6.
Fig. 6. Color-opponent mechanisms produced by our model with macular pigment filtering set to 100% of max (solid lines) and 10% of max (dotted lines), with L:M ratio held constant at 2:1.
Fig. 7. (a) Comparison of data from Burns et al. [16] and (b) model predictions plotted on Judd (1951a) chromaticity diagram. Unique blue and yellow loci (circles) were derived by desaturating a short-wavelength primary while simultaneously adjusting this primary towards unique blue; unique red and green loci (squares) were derived by mixing 460 nm light with a series of long-wavelength lights, ending with the 650 nm match.
Finally, our model also predicts the spectral neutral points of dichromats [19,20]. Because dichromats have only one middle-to-long-wavelength sensitive cone photopigment, the receptive fields responsible for color appearance in dichromats will be the normalized ${\rm{S}}$ versus ${\rm{M}}$ (Protanope) and ${\rm{S}}$ versus ${\rm{L}}$ (Deuteranope) mechanisms produced in the second stage of our model. Assuming ${\rm{M}}$- and ${\rm{L}}$-cone spectra that peak at 530 and 557.25 nm, respectively, and using the same 2 degree stimulus size to calculate macular pigment density, the model predicts zero crossings of 488.6 and 495.4 nm for Protans and Deutans, respectively.
Because there is no L:M ratio to consider, our model predicts that dichromat neutral points will remain relatively stable, showing less variability than the unique green of trichromats. If we normalize the ${\rm{S}}$ versus ${\rm{L}}$ and ${\rm{S}}$ versus ${\rm{M}}$ opponent mechanisms to the same environmental white, any remaining subject-to-subject variability will be the result of differences in the spectral peak of the ${\rm{L}}$ or ${\rm{M}}$ cone photopigment and/or the optical density of the subjects’ lens and macular pigment.
B. Effects of Adaptation
Up to this point, we have not addressed the deviations from linearity that have been demonstrated for both the Blue/Yellow and, to a lesser extent, the Red/Green color-opponent mechanisms [14–17]. As is, our model would predict straight lines connecting unique yellow to unique blue and unique green to unique red, both going through the environmental white used to normalize the cone-opponent mechanisms.
However, because each of our color-opponent mechanisms is made up of not two but three different cones, they are subject to adaptation effects at short and long wavelengths. In our model, the sensation of blueness is driven by both ${\rm{S}}$ and ${\rm{M}}$ cones, while the sensation of redness is driven by both ${\rm{S}}$ and ${\rm{L}}$ cones. This leaves open the possibility that differentially adapting the ${\rm{S}}$ cones at short wavelengths and the ${\rm{L}}$ cones at long wavelengths would shift the location of the constant blue and constant red loci, respectively, in proportion to the amount of adaptation. As the ${\rm{L}}$ cones become increasingly adapted at longer wavelengths, more red light is required to make a match for unique red; by the same token, more blue light will be required to match unique blue as the ${\rm{S}}$ cone becomes increasingly adapted at shorter wavelengths.
We modeled this effect using the two primary techniques used by Burns and colleagues when they showed that the Abney Effect shifts both unique red and unique blue out of alignment with their color-opponent counterparts [16]. To measure desaturated unique green and unique red loci, Burns and colleagues either had their subjects adjust the proportion of a 450 or 470 nm short-wavelength light against a series of long-wavelength lights ranging from 530 to 640 nm, or they used a three-primary technique that desaturated a combination of 460 and 650 nm light with a 510 nm green. To measure desaturated unique blue and unique yellow loci, they used a 571 nm light to increasingly desaturate a variable short-wavelength light. As the amount of 571 nm light in the mixture was increased, the subject was asked to adjust the wavelength of the short-wavelength component until they experienced pure blue or pure yellow.
To test the predictive power of our model, we adjusted the L:M ratio until the unique green produced B/Y mechanism approximately matched the values in the Burns et al. paper (leading to a ratio of 5:1), and then calculated the desaturated hue matches by simulating the experimental conditions previously described. The predictions from our model (right) are shown alongside redrawn figures from Burns et al. (left) in Figs. 7 and 8.
Fig. 8. (a) Comparison of data from Burns et al. [16] and (b) model predictions plotted on Judd (1951a) chromaticity diagram. Unique red and green loci were derived by combining either a 450 nm (squares) or 470 nm (circles) short-wavelength primary with a series of long-wavelength primaries: 530, 550, 570, 580, 590, 610, and 640 nm.
The model predicts the sharp curve in the unique blue loci observed by Burns et al. In our model, this shift is explained by adaptation of the ${\rm{S}}$ cone to the high proportion of short-wavelength light for the points closest to the spectrum locus on the lefthand side. As the proportion and wavelength of the short-wavelength light are changed simultaneously, this quickly releases the adaptation on the ${\rm{S}}$ cone, and the constant hue loci follow a straight line from this point forward.
The model also captures the shift in location and curvature of the unique red and unique green hue loci observed by Burns and colleagues across all the wavelengths tested. In our model, the classic curvature of the unique red loci is the result of increasing ${\rm{L}}$-cone adaptation produced by the long-wavelength hues used in the cancellation experiment. This produces a positive feedback loop where more red light is required to make a match, which further adapts the ${\rm{L}}$-cone, which requires even more red light to make a match, and so on until an equilibrium is reached.
Together, these examples demonstrate that differential adaptation of the ${\rm{L}}$- and ${\rm{S}}$-cone contributions to the (${\rm{S}} + {\rm{M}}$) versus ${\rm{L}}$ and (${\rm{S}} + {\rm{L}}$) versus ${\rm{M}}$ color-opponent mechanisms at long and short wavelengths can explain the curvature of both unique blue and unique red loci in color space.
4. DISCUSSION
A. Seemingly Contrary Evidence
A complete theory of color vision must account for an increasingly complex set of behavioral, psychophysical, anatomical, and physiological data that sometimes seems contradictory. Nowhere is this more relevant than the example of the so-called “unique hues.”
Over the past few decades, the very existence of the unique hues has been called into question [35–38]. In the LGN, the vast majority of cells are tuned to the ${\rm{L}}$ versus ${\rm{M}}$ and ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) directions in color space [39], which do not correspond to the perceptual color-opponent mechanisms that underlie the unique hues. At higher levels of the visual cortex, cells are tuned to every direction in color space [40,41]. Some neural correlates for the unique hues have been described over the years [42,43], but the unique hues remain “one of the central mysteries of color science” [5], and they continue to be “an old problem” for each new generation of vision scientists [44].
In 2015, Wool and colleagues questioned the distinctiveness of the unique hues when they found that these hues are no more salient than any other color when used to perform a behavioral task [35]. However, the expectation that the unique hues, if they have special status, should be more salient than any other colors may come from a misunderstanding that they represent the peaks of the color-opponent mechanisms. As shown in Fig. 4, this is not the case. The special status that the unique hues have been given in color science makes it easy to mistake them for some sort of “super hues” that should be more obvious than so-called “reducible” hues. But the unique hues are nothing more than the product of a single color-opponent system firing on its own when the other system is nulled. If anything, the saliency of the unique hues is predicted to be weaker than colors formed by two different opponent mechanisms firing in unison.
Another potential argument against the unique hues that relies on their “special” status comes from hue scaling experiments [36,45], where a factor analysis identified not four, but seven systematic factors that account for a variation in hue scaling across a narrow range of hues. Like the existence of neurons in the visual cortex that are tuned to many directions in color space, these findings may be interpreted to suggest there is nothing inherently unique about red, green, blue, and yellow. But just as the existence of two cone-opponent mechanisms does not negate the fact that our vision is, at a more fundamental level, based on three classes of cones, the existence of higher-level cortical mechanisms tuned to multiple color directions does not negate the fact that color perception is, at a more fundamental level, based on two color-opponent systems underlying four unique hues.
Finally, the fact that unique red and unique green are not co-linear in color space has been used to bolster arguments that linear color-opponent models cannot be correct. In 2018 [13], Conway and colleagues wrote that “the four chromatic unique hues cannot be related in a simple way to cone-opponent stages of processing.” However, as we have shown, a color-opponent model that combines cone signals as (${\rm{S}} + {\rm{M}}$) versus ${\rm{L}}$ for B/Y and (${\rm{S}} + {\rm{L}}$) versus ${\rm{M}}$ for R/G permits a simple neurobiological explanation for why the Red versus Green and Blue versus Yellow axes are curved in color space [14–17]. In Section 3.B, we show that a linear combination of cone-opponent mechanisms can produce accurate non-linear loci for unique green and unique red (as well as unique blue and unique yellow) if we account for the differential adaptation of the ${\rm{L}}$ and ${\rm{S}}$ cones at long and short wavelengths.
B. Neural Correlates of Unique Hues
Our goal is to develop a theory that explains the neurobiological basis of the first stages of color processing in the human visual system. The model presented here represents a step towards understanding the biological mechanisms that connect the cones to color appearance. It shows that cone-opponent signals can be normalized and recombined into color-opponent mechanism with appropriate spectral positions and variability of the unique hues without resorting to post hoc adjustments. However, the neural circuity underlying this computational model is still in open question.
${\rm{L}}$ versus ${\rm{M}}$ and ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) cone-opponent signals are carried by the midget- and small-bistratified ganglion cells of the retina [46–48], which project to the parvocellular (P) and koniocellular (K) layers of the LGN of both macaque [39] and marmoset [49]. Thus, the multistage models put forward by DeValois and Devalois and Stockman and Brainard propose that the outputs of these ganglion cells are combined somewhere in the visual cortex to produce neurons with the color-opponent tuning described above [10,12]. However, no location in cortex has been identified where the appropriate interactions between signals from the ${\rm{P}}$ and ${\rm{K}}$ layers of the LGN could take place. Moreover, individuals with mutations in the GRM6 gene that prevent signaling between S-cones and blue cone bipolar cells have no ${\rm{S}}$-cone inputs to small bistratified cells. If small bistratified cells were the basis for the ${\rm{S}}$-cone signals of the color-opponent mechanisms, individuals with such mutations should be tritanopes; however, they have perfectly normal color vision [50].
Our group has proposed an alternate possibility: that color-opponent signals are created in the retina, arriving at the LGN fully formed [51–53]. According to this theory, the ${\rm{L}}$ versus ${\rm{M}}$ signal is created by lateral inhibition through HI horizontal cells as held by conventional theories. However, the ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) signal is created by lateral inhibition from HII horizontal cells [54], and the color-opponent signals are created by “injecting” this ${\rm{S}}$ versus (${\rm{L}} + {\rm{M}}$) signal into ${\rm{L}}$ versus ${\rm{M}}$ midget bipolar cells through a sign-reversed GABAergic synapse between HII horizontal cells and midget bipolar cells [55]. This idea benefits from physiological evidence that there is a known HII horizontal cell to midget bipolar cell feed-forward synapse present beneath the ${\rm{L}}$, ${\rm{M}}$, and ${\rm{S}}$ cone pedicles in the primate retina [56].
This would create a small sub-population of midget ganglion cells that carry the color-opponent signals required by our model, leaving the rest of the pure ${\rm{L}}$ versus ${\rm{M}}$ midget ganglion cells free to mediate high-acuity black and white vision [55,57]. One advantage of this theory is that it bypasses the mGluR6 receptor required for ${\rm{S}}$-ON input to the small-bistratified ganglion cell, explaining why individuals with a mutation that knocks out this receptor still have normal color vision [50]. It also explains why the ability to distinguish a red–green grating from black-and-white distractors disappears at an acuity of approximately 10–12 cycles per degree [58]—far lower acuity than the 28–30 cycles per degree possible if all ${\rm{L}}$ versus ${\rm{M}}$ midget ganglion cells were responsible for our sensations of red and green.
The idea that our unique hues are served by a small population of midget ganglion cells with ${\rm{S}}$-cone inputs seems consistent with recording from the parvocellular layers of the primate LGN that have revealed a small population of cells with cone inputs corresponding to the color-opponent mechanisms described here [39,59]. A future direction of our research is to test the hypothesis of a retinal origin for the unique hues by searching for retinal ganglion cells with (${\rm{S}} + {\rm{L}}$) versus ${\rm{M}}$ and (${\rm{S}} + {\rm{M}}$) versus ${\rm{L}}$ inputs in an ex vivo preparation of primate retina, as was used to identify the previously elusive ${\rm{S}}$-OFF midget ganglion cells [60].
Because of its extraordinary explanatory and predictive powers, color-opponent theory is one of the most enduring theories in neuroscience. It explains why we have four irreducible hue sensations, why there is no reddish green or bluish yellow, and why red–green color deficient individuals lose both red and green sensations when they lose one class of cones. Here we show how fundamental features of human biology—the long-term adaptation of color vision to white [61], the variability in cone ratio [62], and differential adaptation of the cones—when applied to color-opponent mechanisms that combine known components of the visual system, can explain the locations and variability of the unique hues.
Funding
Research to Prevent Blindness; National Eye Institute (P30EY001730, RO1EY027859).
Disclosures
The authors declare no conflicts of interest.
Data availability
No data were generated or analyzed in the present research.
REFERENCES
1. H. von Helmholtz, Handbuch der Physiologischen Optik (Leopold Voss, 1896).
2. E. Hering, “Zur Lehre vom Lichtsinne,” in Sechs Mittheilungen an die Kaiserliche Akademie der Wissenschaften in Wien (1878).
3. L. M. Hurvich and D. Jameson, “An opponent process theory of color vision,” Psychol. Rev. 64, 384–404 (1957). [CrossRef]
4. R. L. DeValois, I. Abramov, and G. H. Jacobs, “Analysis of response patterns of LGN cells,” J. Opt. Soc. Am. 56, 966–977 (1966). [CrossRef]
5. J. D. Mollon and G. Jordan, “On the nature of unique hues,” in John Dalton’s Colour Vision Legacy, C. Dickenson, L. Murray, and D. Carden, eds. (Taylor & Francis, 1997), pp. 381–392.
6. D. H. Krantz, “Color measurement and color theory: II. Opponent-colors theory,” J. Math. Psychol. 12, 304–327(1975). [CrossRef]
7. J. Larimer, D. H. Krantz, and C. M. Cicerone, “Opponent-process additivity—I: Red/green equilibria,” Vis. Res. 14, 1127–1140 (1974). [CrossRef]
8. C. M. Cicerone, D. H. Krantz, and J. Larimer, “Opponent-process additivity—III: Effect of moderate chromatic adaptation,” Vis. Res. 15, 1125–1135 (1975). [CrossRef]
9. J. Larimer, D. H. Krantz, and C. M. Cicerone, “Opponent process additivity—II. Yellow/blue equilibria and nonlinear models,” Vis. Res. 15, 723–731 (1975). [CrossRef]
10. R. L. DeValois and K. K. DeValois, “A multi-stage color model,” Vis. Res. 33, 1053–1065 (1993). [CrossRef]
11. S. L. Guth, “Model for color vision and light adaptation,” J. Opt. Soc. Am. A 8, 976–993 (1991). [CrossRef]
12. A. Stockman and D. Brainard, “Color vision mechanisms,” in OSA Handbook of Optics, M. Bass, ed., 3rd ed. (McGraw-Hill, 2010), pp. 11.11–104.
13. B. R. Conway, R. T. Eskew, P. R. Martin, and A. Stockman, “A tour of contemporary color vision research,” Vis. Res. 151, 2–6 (2018). [CrossRef]
14. F. L. Dimmick and M. R. Hubbard, “The spectral location of psychologically unique yellow, green, and blue,” Am J. Psychol. 52, 242–254 (1939). [CrossRef]
15. F. L. Dimmick and M. R. Hubbard, “The spectral components of psychologically unique red,” Am. J. Psychol. 52, 348–353 (1939). [CrossRef]
16. S. A. Burns, A. E. Elsner, J. Pokorny, and V. C. Smith, “The Abney effect: Chromaticity coordinates of unique and other constant hues,” Vis. Res. 24, 479–489 (1984). [CrossRef]
17. J. S. Werner and B. R. Wooten, “Opponent chromatic mechanisms: relation to photopigments and hue naming,” J. Opt. Soc. Am. 69, 422–434 (1979). [CrossRef]
18. B. E. Schefrin and J. S. Werner, “Loci of spectral unique hues throughout the life span,” J. Opt. Soc. Am. A 7, 305–311 (1990). [CrossRef]
19. C. M. Cicerone, A. L. Nagy, and J. L. Nerger, “Equilibrium hue judgements of dichromats,” Vis. Res. 27, 983–991 (1987). [CrossRef]
20. G. L. Walls and G. G. Heath, “Neutral points in 138 protanopes and deuteranopes,” J. Opt. Soc. Am. A 46, 640–649 (1956). [CrossRef]
21. J. Carroll, M. Neitz, and J. Neitz, “Estimates of L:M cone ratio from ERG flicker photometry and genetics,” J. Vis. 2(8), 531–542 (2002). [CrossRef]
22. J. Neitz and M. Neitz, “The genetics of normal and defective color vision,” Vis. Res. 51, 633–651 (2011). [CrossRef]
23. J. Carroll, C. Bialozynski, P. Summerfelt, M. Neitz, and J. Neitz, “Estimates of L:M cone ratios from ERG flicker photometry and genetics,” Invest. Ophthalmol. Vis. Sci. 41, S808 (2000).
24. B. R. Hammond, B. R. Wooten, and D. M. Snodderly, “Individual variations in the spatial profile of human macular pigment,” J. Opt. Soc. Am. A 14, 1187–1196 (1997). [CrossRef]
25. A. Stockman, “Colour and vision research laboratory and database,” (1995–2019), https://www.cvrl.org.
26. G. Wyszecki and W. S. Stiles, Color Science: Concepts and Methods, Quantitative Data and Formulae (Wiley, 1967), p. 628.
27. L. M. Hurvich, Color Vision (Sinauer Associates, 1981).
28. D. Rezeanu, “ColorOpponentModel,” GitHub (2022), https://bit.ly/3xZG1Ir.
29. G. Ekman, “Weber’s law and related functions,” J. Psychol. 47, 343–352 (1959). [CrossRef]
30. J. D. Mollon and G. Jordan, “Is unique yellow related to the relative numerosity of L and M cones?” Invest. Ophthalmol. Vis. Sci. 36, S189 (1995).
31. G. Jordan and J. D. Mollon, “Rayleigh matches and unique green,” Vis. Res. 35, 613–620 (1995). [CrossRef]
32. I. Abramov and J. Gordon, “Seeing unique hues,” J. Opt. Soc. Am. A 22, 2143–2153 (2005). [CrossRef]
33. R. Kuehni, “Variability in unique hue selection: A surprising phenomenon,” Color Res. Appl. 29, 158–162 (2004). [CrossRef]
34. R. G. Kuehni, “Determination of unique hues using Munsell color chips,” Color Res. Appl. 26, 61–66 (2001). [CrossRef]
35. L. E. Wool, S. J. Komban, J. Kremkow, M. Jansen, X. Li, J.-M. Alonso, and Q. Zaidi, “Salience of unique hues and implications for color theory,” J. Vis. 15(2), 10 (2015). [CrossRef]
36. K. J. Emery and M. A. Webster, “Individual differences and their implications for color perception,” Curr. Opin. Behav. Sci. 30, 28–33 (2019). [CrossRef]
37. J. Bosten and A. Boehm, “Empirical evidence for unique hues?” J. Opt. Soc. Am. A 31, A385–A393 (2014). [CrossRef]
38. B. A. Saunders and J. Van Brakel, “Are there nontrivial constraints on colour categorization?” Behav. Brain Sci. 20, 167–179 (1997). [CrossRef]
39. A. M. Derrington, J. Krauskopf, and P. Lennie, “Chromatic mechanisms in lateral geniculate nucleus of macaque,” J. Physiol. 357, 241–265 (1984). [CrossRef]
40. P. Lennie, J. Krauskopf, and G. Sclar, “Chromatic mechanisms in striate cortex of macaque,” J. Neurosci. 10, 649–669 (1990). [CrossRef]
41. K. S. Bohon, K. L. Hermann, T. Hansen, and B. R. Conway, “Representation of perceptual color space in macaque posterior inferior temporal cortex (the V4 complex),” eNeuro 3 , ENEURO.0039-16.2016 (2016). [CrossRef]
42. C. M. Stoughton and B. R. Conway, “Neural basis for unique hues,” Curr. Biol. 18, R698–R699 (2008). [CrossRef]
43. L. Forder, J. Bosten, X. He, and A. Franklin, “A neural signature of the unique hues,” Sci. Rep. 7, 42364 (2017). [CrossRef]
44. A. Valberg, “Unique hues: an old problem for a new generation,” Vis. Res. 41, 1645–1657 (2001). [CrossRef]
45. K. J. Emery, V. J. Volbrecht, D. H. Peterzell, and M. A. Webster, “Variations in normal color vision. VII. Relationships between color naming and hue scaling,” Vis. Res. 141, 66–75 (2017). [CrossRef]
46. J. Crook, M. Manookin, O. Packer, and D. Dacey, “Horizontal cell feedback without cone type-selective inhibition mediates “red-green” color opponency in midget ganglion cells of the primate retina,” J. Neurosci. 31, 1762–1772 (2011). [CrossRef]
47. D. M. Dacey and B. B. Lee, “The blue-ON opponent pathway in primate retina originates from a distinct bistratified ganglion cell type,” Nature 367, 731–735 (1994). [CrossRef]
48. H. Kolb and D. Marshak, “The midget pathways of the primate retina,” Doc. Ophthalmol. 106, 67–81 (2003). [CrossRef]
49. K. A. Percival, P. R. Jusuf, P. R. Martin, and U. Grunert, “Synaptic inputs onto small bistratified (blue-ON/yellow-OFF) ganglion cells in marmoset retina,” J. Comp. Neurol. 517, 655–669 (2009). [CrossRef]
50. T. P. Dryja, T. L. McGee, E. L. Berson, G. A. Fishman, M. A. Sandberg, K. R. Alexander, D. J. Derlacki, and A. S. Rajagopalan, “Night blindness and abnormal cone electroretinogram ON responses in patients with mutations in the GRM6 gene encoding mGluR6,” Proc. Natl. Acad. Sci. USA 102, 4884–4889 (2005). [CrossRef]
51. J. Neitz and M. Neitz, “Evolution of the circuitry for conscious color vision in primates,” Eye—London 31, 286–300 (2017). [CrossRef]
52. B. P. Schmidt, M. Neitz, and J. Neitz, “The neurobiological explanation for color appearance and hue perception,” J. Opt. Soc. Am. 31, A195–A207 (2014). [CrossRef]
53. B. P. Schmidt, P. Touch, M. Neitz, and J. Neitz, “Circuitry to explain how the relative number of L and M cones shapes color experience,” J. Vis. 16(8), 18 (2016). [CrossRef]
54. O. Packer, J. Verweij, P. Li, J. L. Schnapf, and D. M. Dacey, “Blue-yellow opponency in primate S cone photoreceptors,” J. Neurosci. 30, 568–572 (2010). [CrossRef]
55. S. S. Patterson, M. Neitz, and J. Neitz, “Reconciling color vision models with midget ganglion cell receptive fields,” Front. Neurosci. 13, 865 (2019). [CrossRef]
56. C. Puller, S. Haverkamp, M. Neitz, and J. Neitz, “Synaptic elements for GABAergic feed-forward signaling between HII horizontal cells and blue cone bipolar cells are enriched beneath primate S-cones,” PLoS One 9, e288963 (2014). [CrossRef]
57. D. Rezeanu, M. Neitz, and J. Neitz, “How we see black and white: The role of midget ganglion cells,” Front. Neuroanat. 16, 944762 (2022). [CrossRef]
58. A. Neitz, X. Jiang, J. A. Kuchenbecker, N. Domdei, W. Harmening, H. Yan, J. Yeonan-Kim, S. S. Patterson, M. Neitz, and J. Neitz, “Effect of cone spectral topography on chromatic detection sensitivity,” J. Opt. Soc. Am. A 37, A244–A254 (2020). [CrossRef]
59. C. Tailby, S. G. Solomon, and P. Lennie, “Functional asymmetries in visual pathways carrying S-cone signals in macaque,” J. Neurosci. 28, 4078–4087 (2008). [CrossRef]
60. S. S. Patterson, J. A. Kuchenbecker, J. R. Anderson, A. S. Bordt, D. W. Marshak, M. Neitz, and J. Neitz, “An S-cone circuit for edge detection in the primate retina,” Sci. Rep. 9, 11913 (2019). [CrossRef]
61. J. Neitz, J. Carroll, Y. Yamauchi, M. Neitz, and D. R. Williams, “Color perception in mediated by a plastic neural mechanism that is adjustable in adults,” Neuron 35, 783–792 (2002). [CrossRef]
62. H. Hofer, J. Carroll, J. Neitz, M. Neitz, and D. R. Williams, “Organization of the human trichromatic cone mosaic,” J. Neurosci. 25, 9669–9679 (2005). [CrossRef]
