Opponent process is a color theory that states that the human visual system interprets information about color by processing signals from cones and rods in an antagonistic manner (source).

What is the advantage of opponent color against RGB color? For example, in object recognition or edge detection?

Can we say that it provides some level of invariance to changes in brightness?


RGB color is a more direct representation of the "raw input" received by the (human) eye, since the three types of cone cell have responsivity spectra that roughly correspond to red, green, and blue light:

Human cone response curves

Opponent color decomposes color into three "dimensions" of opposing properties: dark or light, yellow or blue, and red or green. This provides some brightness invariance, since changing the intensity of light will change the brightness but not the yellow/blue or red/green balance of the color, even though the stimulation level of all three types of cone cell will change.

One could speculate that opponent color perception is more "meaningful" in real-world environments. The dark/light axis represents the overall amount of light, the yellow/blue axis indicates which end of the visible spectrum is predominant, and the red/green axis provides a finer distinction for low-frequency light.

  • $\begingroup$ "...even though the stimulation level of all three types of cone cell will change": The stimulation levels change, but the ratios of r/g/b/ cone activation remain the same under varying luminance, regardless of the opponent channels. Hence opponency is not necessary for color constancy. $\endgroup$ – AliceD Jan 9 '15 at 1:01
  • $\begingroup$ and why would 'end-of-spectrum' information of the yb axis be helpful? $\endgroup$ – AliceD Jan 9 '15 at 1:02
  • $\begingroup$ @ChrisStronks Of course color opponency is not the only way to achieve color constancy, I'm not sure what your point is. $\endgroup$ – augurar Jan 9 '15 at 3:25
  • $\begingroup$ @ChrisStronks The "end of spectrum" information is helpful in finding fruits or young leaves among other foliage. This is thought to be why trichromacy evolved in primates (most other mammals have only two cones). $\endgroup$ – augurar Jan 9 '15 at 3:28
  • $\begingroup$ Trichromacy helps in fruit recognition for sure, but the added color is either red or green, the dynamic frequency range wasn't changed much when trichromats appeared. See the benefit of trichromats and fruit in this answer: biology.stackexchange.com/questions/24481/… $\endgroup$ – AliceD Jan 9 '15 at 3:33

The first question, namely what is the advantage of color opponency 'against' RGB, is technically incorrect. The opponent system (Red/Green; Blue/Yellow and brightness channels) is physiologically situated in the neuronal retinal layer and higher visual structures such as the lateral geniculate nucleus and receives input from the RGB system (photoreceptors in the retina: Red, Green, Blue cones).

enter image description here

Then the question becomes "what is the advantage of adding the opponent channels on top of the RGB system? The advantage of opponency may in fact not be the negative opponent effect so often stressed in the Hering model (and shown in the figure: e.g., blue suppressing yellow etc). The advantage lies in the finding that activity of one opponent pair increases activity of the opponent color in adjacent areas in the retina. This effect results in a sharpening of color contrasts between the opponent colors (Hurvich and Jameson, 1957). Indeed, the poor spatial color-contrast, as provided by the sparsely placed cones in the retina, is enhanced by this mechanism. Especially the blue cones are present in isolated patches in the retina. Because one color in an opponent pair increases the other's activity directly adjacent to it in the retina, the resolution of the blurry retinal image is greatly increased. Hence, as you rightfully say, edge detection and in particular color contrast perception is enhanced.

Object recognition, as you mention in your question, is a process that is facilitated by the visual system through many stages including the higher visual cortices. Enhanced color contrast will likely facilitate object perception, but indirectly since color contrasting makes up just a small portion of a much bigger and complicated neural system devoted to object recognition, namely the visual 'ventral' stream (Ishai et al., 1999).

Lastly you ask whether opponent colors may facilitate invariance to brightness. Assuming you mean color invariance under varying luminance (referred to as color constancy) the answer is no. The RGB cone system can deal with this easily, as the ratio of cone activation yields enough information to code the hue under various luminances. If a certain object reflection activates twice as many red cones than green cones, it will also activate the same ratio when the luminance is increased (Mather, 2006 - Chapter 12).

  • $\begingroup$ Technically, the contrast effect is due to spatial opponency rather than color opponency. $\endgroup$ – augurar Jan 9 '15 at 3:45
  • $\begingroup$ @augurar - There are many other opponent mechanisms, and you are referring to the center-surround receptive fields of retinal ganglion cells that increase spatial contrast. I am talking about color contrast, which is a different mechanism, namely color opponency (the topic of the question) $\endgroup$ – AliceD Jan 9 '15 at 4:18
  • $\begingroup$ @AliceD, could you expand on the last point please - why do you think the opponent process does not facilitate color constancy? You're saying that 'RGB cone system can deal with it easily by calculating the ratio of cone activation', - but the opponent process is exactly that, the calculation of the ratio between cone activations. It just so happens that our particular visual system does this via the R-G, B-Y, Brightness variables. $\endgroup$ – lakesare Apr 13 '20 at 7:25

The opponent process is indeed useful for color constancy.

Opponent process maps RGB (or LMS) coordinates to YB-RG-WhiteBlack coordinates:

enter image description here

RG axis, for example, is calculated by the simple algebraic R - G (or L - M).
If the brightness is reduced, there will be fewer photons falling on both the R cones and the G cones.
The WhiteBlack value will change:

R + G + B = ?
8 + 3 + 0 = 11 →
7 + 2 + 0 = 9 →
6 + 1 + 0 = 7 --- notice WhiteBlack value changes 

The difference between R and G will remain the same:

R - G = ?
8 - 3 = 5 →
7 - 2 = 5 →
6 - 1 = 5 --- notice RG value stays the same

This ensures color constancy - the visual system has a way to determine that the color of the object didn't change, only the illumination did.

Here is a related quote from Vision Science: Photons to Phenomenology by Stephen Palmer:

The opponent representation is more useful because it helps determine which differences between adjacent retinal regions result from changes in the level of illumination (such as when a shadow falls across a surface) and which result from changes in spectral reflectance (such as when a surface is painted two different colors). Changes in the amount of illumination will generally just raise or lower the overall amount of light (that is, change the output of the B1/Wh system) while the chromatic balance remains constant (that is, the outputs of the R/G and B/Y systems are invariant).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.