Blog  |   Puzzles  |   Books  |   About

Emergent Orange

Emergent Orange Example StepsEmergent Orange describes the orange hue that is produced when you average together a bunch of randomly selected digital photographs.

The illustration shows 5 different sets of photos (randomly selected from Flickr) accumulating over successive rows. The first row is a single image. Then 2 images, 5 images, 25 images, and 100 images in the bottom row. I have cranked up the saturation to reveal the orange shift (unprocessed averages tend to look like dirt or milk-chocolate).

I stumbled across this effect in 2006, playing with Flickr, and have blogged about it a few times. Other digital artists who use the same averaging technique have also observed the effect. The reasons why it happens are not yet entirely clear, but I suspect it has something to do with chemistry and physics. Interestingly the same effect occurs with collections of human-generated synthetic, abstract art (not photos), such as fractals.

Over thanksgiving break I wrote an informal paper describing my findings to date, in hopes of attracting some brighter minds to the topic. Perhaps you are one of those minds? You’ll find my paper here.

If you prefer pretty pictures to words, here’s a Flickr set containing some of my image averaging experiments.


4 Responses to “Emergent Orange”

  1. Goddess Finchy Says:

    You have noted the effect known to artists by the technical expression “turning everything to shit.” Invariably, if you mix more than two colors, you’re going to have a third really crappy color. The more colors you add, even from different ends of the spectrum, or complementary color, the crappier the result, usually the result is a really ugly brown. White is the combination of all colors of light, black is the absence of light, which is physics. However, when one is dealing with pigment (represented by pixels in an image) black and white are both are colors, which is chemistry.

    Totally useless fact: When you eat food, it usually comes out brown, unless you eat flubber, which comes out green (even it was blue.) I know because it happened to my grandson.

    http://colormatters.com/

  2. jbum Says:

    One of the reasons I tend to refer to this as orange (even though the unnormalized color is brown) is that the word “brown” is overloaded – it tends to be used to refer to a large number of unsaturated hues of varying compositions.

    My main observation here is that the resultant hue (in the Hue, Saturation, Brightness sense) is at a fairly specific spot on the hue ring – generally within a 10-degree area. It’s a fairly specific *kind* of brown, which becomes orange when the saturation is cranked up.

    Interestingly, since you mention paints, I should mention that I also ran some tests on the available swatches of commercial paint catalogs (e.g. Sherwin Williams etc.) and ran into similar results.

  3. Michael J Swart Says:

    Cool!, Here’s more confirmation!
    http://www.flickr.com/photos/mjswart/8549909110/
    The image I linked to is an average of 100 illustrations I create for my blog here:
    http://michaeljswart.com/browse-articles-by-illustration/

    The average is definitely a grey/brown And the hue definitely leans toward orange. The only criteria I used for colours is what looks good to me as well as some variety across illustrations.

  4. Mick Mengucci Says:

    Hi, nice paper! To my perspective this is an evidence of the fact that usually colors considered ‘warm’ have a higher value of Value (in HSV color system) or Brightness (in HSB). I found this during this research (http://ctp.di.fct.unl.pt/~sc/SeeThroughSound/) where I programmed a tool able to sonificate the color data of digital images for blind users. Usually the value of V captured from the real world by cameras is much higher for warm colors than for cool colors. Even if they “look” equally bright to human eye. So when you average colors the warm part of color spectrum prevails, and that emerging orange could be exactly the color with higher intrinsic V. The fact that when you average images created digitally this does not happen confirms this theory. In fact here the Value and Saturation are at maximum value for each color and you just choose random Hues, right? that is why there is no emerging color, they all have the same initial Value.
    I hope it helps. Keep me updated if you get confirmation or not about this theory as it could turn out an interesting issue in research.