Biologists and linguists alike have long puzzled over the universal human tendency to see and name colors as a combination of four "pure" hues — red, yellow, green and blue.
Most people and most cultures do not perceive red as a combination of orange or purple, for example, whereas they naturally perceive orange as a combination of yellow and red. We also perceive some colors as opposites of one another: red is the opposite of green; blue the opposite of yellow. We would never describe a color as reddish-green or bluish-yellow.
Biologists have tried and failed to reconcile these four primary opposing colors with the three kinds of cone cells in our eye that detect different wavelengths of light. Similarly, our subjective experience of color can't be explained by the network of neurons that process vision in the brain. Yet, as painstakingly documented by linguist Paul Kay and the late anthropologist Brent Berlin at UC Berkeley, dozens of human languages, including many unwritten ones, categorize color based on these four hues, hinting at a unique aspect of human perception.
"We tend to think of red as just red, not as a mixture," said Berkeley postdoctoral fellow Alexander Belsten. "This same unique property holds for green, blue and yellow: each appears perceptually pure rather than as a mixture of neighboring hues."
Belsten and Berkeley neuroscientist Bruno Olshausen have now come up with a theory that explains this conundrum. They demonstrate that the natural world — as distinct from the more colorful human-created world — displays a restricted palette of colors that the human brain, for the sake of simplicity, represents as combinations of only four pure colors. Basically, a combination of just four opposing hues — red vs green and blue vs yellow — provides the simplest way to encode the range of colors found in nature.
"Think about this like north, south, east and west. If you say you're going north, that means you're not going south," said Olshausen, director of Berkeley's Redwood Center for Theoretical Neuroscience and professor of optometry and vision science. "That's what's kind of striking about this study. It captures exactly this idea of opponency that physiologists described back in the 19th century, which is that blue and yellow appear to be opposites and that red and green appear to be opposites. This has puzzled color vision researchers for a long time."
The new theory offers "a possible resolution between two historically competing accounts of color vision that emerged around the same time, and were vigorously debated, in the late nineteenth century: Hermann von Helmholtz's theory that color vision begins with three receptor types vs. Ewald Hering's theory of the four unique hues and their opponent nature," said Belsten, who is first author of a paper about the theory published July 14 in the Journal of the Optical Society of America A. "We now know that both are valid accounts for how we see color. The resolution is that while the former speaks to physiological sensory mechanisms in the retina, the latter describes a psychological basis for describing our subjective experience of color … that is grounded in the structure of the natural visual environment."
It's likely, Olshausen said, that our primate relatives, which also have three types of color-sensing cones in the eye, perceive hues in a similar way — as unique, opponent hues.
Color theory
Ewald Hering proposed his opponent-process theory of color vision in 1878 and referred to the four primary colors as "urfarben," German for "primitive colors." It is the basis of color theory today, including color wheels used by designers worldwide.
"Yet their special status remains a mystery," Belsten said. "No known property of light — the cone photoreceptors in the eye or neural representations in the brain — can explain why these four hues, rather than some other set, should occupy this privileged position."
Though still debated among linguists, the studies by Kay and Berlin demonstrated the universality of these privileged colors. They showed that, though cultures ranged widely in the number of words they used to differentiate colors, they all named subsets of the basic colors white, red, yellow, green, blue and black.
Several years ago, Olshausen, who studies how the natural visual environment shapes our perception, asked Belsten to take another look at the problem, employing recently developed datasets of the color or spectral distribution of natural scenes. They used this calibrated data, compiled in part to ground-truth satellite images of Earth, to simulate what the cones in the human eye perceive when confronted with images of nature, representative of a time before humans created more saturated pigments.
While modern color display technology commonly produces colors as some combination of red, green and blue, the three types of color photoreceptors in the human eye, called cone cells, do not directly measure those colors. The short-wavelength cone responds most strongly to blue, but the medium- and long-wave cones overlap in sensitivity and respond strongest to different shades of orange or yellow.
When the researchers simulated how the three cones respond to the spectral distribution of the natural world, they found that most of the variation is in brightness — light to dark — and that most of the world's pixels are a colorless gray. Plotting the color-only response, it became clear that the eye's sensitivity to the colors of the natural world is not distributed equally across all hues, but peaks in three regions: most prominently red, but also yellow-green and blue-green. Except for red, the hues in nature are overwhelmingly unsaturated: pale blue skies and light yellow-green vegetation contrast with the more saturated red hues in tree trunks, rocks and soil.
"One major finding from the data is that the color distribution in the environment is very asymmetric and non-uniform," Olshausen said. "Nobody's ever looked at this before because the availability of these very large data sets is relatively new. You need to have millions and millions of pixels to start to see the dominant colors, because some color pixels are only one in 100,000 or one in a million."
Sparse coding
Belsten then asked how the brain would represent this color distribution to minimize the activity of neurons needed to encode the natural world — a representation called sparse coding that was propounded in the 1960s by the late Berkeley vision scientist Horace Barlow. Olshausen has since shown that the brain uses sparse coding to more efficiently represent a large range of sensory information.
"People have found that neurons in the brain seem to be organized to form a sparse representation of different sensory modalities, from gray-level images and sound to touch and smell, and now color," Olshausen said.
The researchers discovered that while three types of neurons — red, yellow-green and blue-green — could easily encode all colors in nature in combinations of two, four types of neurons are better. Four — which just happen to fall out of the theory as red, yellow, green and blue — not only allow a simple representation of any hue as a combination of two of the four, but also allow mutually exclusive color pairs — that is, red versus green and blue versus yellow. These "instantaneous representations by a set of neurons in the brain" can also serve as the basis of color memory, he said.
"Humans do it this way because neurons can now describe the world in terms of something and its opposite. Red is not green, just like north is not south. You don't have this parsimonious representation in terms of opponent colors if you have only three colors to encode the world," Olshausen said. "What I'm really excited about is that we have an explanation now for the first time. All these pieces just fall together. It's almost too good to be true."
"Sparse coding inference introduces mutual exclusivity … and yields a color representation that mirrors the phenomenology of color perception," Belsten added.
Belsten's findings are similar to those found 10 years earlier by another of Olshausen's postdoctoral fellows, E. Paxon Frady, who employed images he found on the internet that were not precisely color calibrated and which included human-made objects. Belsten, Frady and Olshausen are co-authors of the recent paper. Olshausen is also a member of Berkeley's Helen Wills Neuroscience Institute.
The work was funded by the Air Force Office of Scientific Research and is part of the Multi-University Research Initiative that also found a way to produce a color, olo , that had never before been seen by humans.
