Edwin H. Land's Retinex Theory

Regrettably, the ordinary reading that we have, containing the key article by Land, is no longer available in book form. There are published documents, of course. The red and white study is from Scientific American, May, 1959. The retinex theory work with the Mondrian patterns is described in Scientific American, December, 1977.

The term, "retinex" was coined by Land to indicate his belief that full color perception, with color constancy, involved all levels of visual processing, from the retina to the visual cortex. The term is meant to capture this by combining the word "retina" with "cortex."

Here's the main argument behind retinex theory as shown in the Mondrian demonstrations:

Land used 3 different sources of narrow wavelength illumination to shine on the Mondrian color display. The wavelengths used were 630 nm (long), 530 nm (medium), and 450 nm (short). If only one of those was shined on the display, no color variation could be seen, just various levels of light and dark. One MIGHT claim to see color in one of these displays, but only as a kind of covering film. The colored squares might all look like they were degrees of blue (or red or green), as if looking through blue (or red or green) sunglasses at a black and white scene. If the covering haze seems colored, what's behind it is not. Therefore I stress that one does not see variation of colors (hues), only light and dark variation in the "Mondrian" display illuminated by one monochromatic (single wavelength) light.

If one shines all 3 illuminators at once on the display, it is essentially shining white light on them because all 3 light primaries are shining on the display at once. Therefore, even though there are only 3 wavelengths present, and all other wavelengths are absent, these represent enough of the spectrum for color to show up.

As long as some illumination at each of the 3 wavelengths is illuminating the "Mondrian," the colors of the colored rectangles can be seen. The actual amounts of each wavelength can vary a great deal without affecting the colors seen. This is the basis of Land's demonstration. Given this, he can vary the amount of light in each source of illumination. That will, in turn, affect the amount of light at that wavelength bouncing off of any colored area in the display.

What he did, then, was to vary the quantity of illumination at each wavelength so that he could have exactly the same amount of long, medium, and short wavelength light coming from ANY colored area (one trick being, of course, that these identical quantities could never come from different colored areas at the same time).

In our illustrations of 3 conditions, we show relative quantities of light (not any real units) at each of the 3 wavelengths of illumination striking 3 different colored areas on the "Mondrian." For the light shining on the RED area, Land illustrated with 7.4 units of long wavelength illumination, 18.8 units of medium, and 6.1 units of short wavelength illumination. Reflected off of the red area are 5.8 units of long wavelength light, 3.2 units of medium wavelength light and 1.6 units of short. For the red area then, we can say that the reflectance of the long is 5.8 / 7.4, of the medium is 3.2 / 18.8, and of the short is 1.6 / 6.1. Together we call these spectral reflectance.

By manipulating the quantities in each illumination light, Land showed that he could make any colored area have 5.8 units of long wavelength light, 3.2 units of medium, and 1.6 units of short --- and you would still see the "original" or "real" color of the Mondrian rectangular area. Therefore, knowing all the wavelengths coming to your eye from a given colored area would NOT tell you what color a person would see in looking at that area.

What's important is what's happening to the neighbors at the same time.

Here's a quiz to see if you are following and can apply what you should know very well by now: In the diagram where the illumination is falling on the BLUE AREA of the "Mondrian," how much short, medium, and long wavelength light is bouncing off of the red area and headed toward an observer's eye? Hint: You already should have calculated the relevant reflectances.

Land's "Retinex" theory of color vision is meant to account for the usual facts of color vision like we have been studying PLUS Color constancy in a natural way. At the core of Land's theory is the acceptance of something like Wallach's views of black - gray - white perception based on RATIOS of light intensity. He believed that each of the 3 kinds of cones in the eye acted as a filter for the scene and resulted in a black and white picture (or light and dark, if you will). What was important was that there be a lightest part of each "picture," a darkest part and intermediate lightness areas. It did NOT matter how light or how dark an area was in terms of amounts of light coming to an observer's eye. All that mattered was the ordering of areas compared to one another in terms of light and dark.

The three gray images I've included with the "Mondrian" set represent the point. Each one represents areas of relative light and dark as filtered by a long, medium, or short wavelength filter. For a red filter, long wavelength light gets through easiest and therefore makes a light area on a black and white print. Wavelengths that are absorbed, and do not get through, therefore leave dark areas on a print. Because I've illustrated each of the 3 different illumination conditions, the lightest and darkest (and intermediate) areas are NOT all the same across pictures. An area that is red, comes through as light if filtered by a red filter, but dark if filtered by either medium or short wavelength filtering.

Land's theory is that the cone - to - cortex systems for each of the 3 wavelength ranges -- short, medium, and long -- results in something like a black and white picture. He argued that color is computed from comparing the relative lightness of a given part of a scene across the 3 "black and white pictures." Thus, an area that is light in the short wavelength filtered black and white picture, but dark in the other two, will be experienced as blue. An area that is light ONLY in the medium filtered black and white picture will be experienced as green, and an area that is light only in the long wavelength filtered "picture," would be experienced as red. An area that is white in all 3 would be white and an area that is black in all 3 would be black. What would magenta be? Yellow? You should be able to look at a set of 3 black and white pictures of the same scene, taken with the 3 different filter conditions, and figure out what color a given area is.

The most recent biography of Land is listed at: Amazon.com

Return to Perception Syllabus