64

A tweet from Lionel Page (@page_eco) showing an optical illusion recently went viral, with the image appearing in many aggregation web-sites:

This is a black and white photograph. Only the lines have colour.

What you “see” is what your 🧠 predicts the reality to be, given the imperfect information it gets.

Optical Illusion

Putting aside the blurring of the distinction between black-and-white and grey-scale, is it true that the areas not marked with a coloured line are truly grey?

  • Comments are not for extended discussion; this conversation has been moved to chat. – Sklivvz Aug 6 at 10:55
85

As Øyvind Kolås, the original creator of the ilusion noted:

The visual experiment that went along with my previous post went viral, the one that went viral was an already lossily compressed JPEG that has also gone through further cycles of viral degradation by scalings and further JPEG recompressions.

He gives a higher quality version:

PNG image from author

It should be easy enough to verify from a colour picker or counting the pixel colours that there is no bleeding of colour in the original. It certainly looks plausible, unlike the JPEG version.

  • 7
    I can tell just by looking in this image. The point you're directly staring at appears grey, but the rest appears coloured. – Chris Cudmore Aug 2 at 19:40
  • 1
    This is an application of dithering for error correction. Starting with an "ideal" color image, you fix certain pixels to grayscale. This introduces "color error" — it makes that local area of the image just a bit redder, or just a bit less green, or whatever. So you take that color error and propagate it over to one of the pixels whose color value hasn't been fixed yet. Notice the skin in the photo above isn't "flesh-tone" — it's frankly "orange," because it's absorbed a lot of reddish error from the gray portions nearby. – Quuxplusone Aug 3 at 15:08
  • @WGroleau It works best for me if I'm reading the text of the post above or below the image, due to the tricks the brain uses to fill in gaps in information, the colored lines get interpreted as filling the whole region. Of course, the longer I stared at things the less well it worked. See also xkcd.com/1080 – Draco18s Aug 3 at 16:59
  • 5
    @ChrisCudmore: Just looking at the points appearing grey doesn't prove they are grey. Inspecting the original with an editor shows the "grey" points are not pure grey, and this answer explains that this is an artifact of JPEG compression, but the effect works even if without that unintentional assistance. – Oddthinking Aug 4 at 7:36
  • 1
    Wow. I'm totally fooled by the "yellow" text across the bottom. I can't tell it's actually white with lines over it no matter how closely I look! ;-) – T.J. Crowder Aug 4 at 13:47
49

Yes.

this is the so-called Munker illusion, similar of White's illusion.

So named after Munker H (1970): "Farbige Gitter, Abbildung auf der Netzhaut und übertragungstheoretische Beschreibung der Farbwahrnehmung" [Chromatic grids, projection to the retina, and translation theory-based description of the color perception].

In psychology books handling human optical perception this is well known.

As an example consider the following higher-res picture in the following picture, with a big grid and two excluded zones. Make the grid a bit finer or increase the viewing distance just a little bit and the illusion gets better:

enter image description here enter image description here

enter image description here enter image description here

(click pictures to enlarge)

Picture in question and compared to monochromatic black grid:

enter image description here enter image description here

In the next version 3.0 of The GIMP, this will be a ready made plugin.

Steps to reproduce:

  1. get colour image
  2. copy into grayscale layer, hide colour layer
  3. create rotated fine raster in yet a newer layer
  4. create selection from raster, hide that layer
  5. copy raster selection from colour layer
  6. paste copied raster colour over grayscale layer
  7. merge visible layers

presto enter image description here enter image description here

  • 1
    Huh, one of these days I should replace that old avatar, or update it to reflect the "new" Skeptics logo - maybe I only need to edit the diagonal lines though! – Oddthinking Aug 2 at 15:17
  • 13
    So people are taking already colored images, converting them to grayscale, then using an illusion to put the color back? Weird. – fredsbend Aug 2 at 15:55
  • 2
    I find it fascinating how, in the monochromatic grid overlay image of the three girls, even the tiny amount of green where the black grid doesn’t completely cover the green grid is enough to make the girl on the right’s t-shirt look completely green. The color-to-greyscale area ratio must be about 1:100 there (including the black grid), and it still works. – Janus Bahs Jacquet Aug 4 at 9:32
  • 7
    @fredsbend: there's a whole subgenre of web development that is all about "how can I have a very cheap image preview that I will later asynchronously replace with a better version" where having a grayscale image and color values at a fraction of the original resolution might be interesting. (Remember progressive jpegs that start out blocky and then get more refined? That kind of thing.) – Ulrich Schwarz Aug 4 at 9:33
  • 2
    I wouldn't call this an "illusion" though. It would be better to describe it as a colour image reproduction method with really bad aliasing artifacts if you look closely. In fact if you look real close at the images on this page, you'll see that they're all "illusions". Even the supposed illusions are actually illusions of illusions. There are no shades of grey on this page, no coloured lines, just small tiny red, green and blue dots. – Ross Ridge Aug 4 at 18:51
35

The human visual system (i.e. eye + brain) processes chroma (e.g red vs green) differently to luminance (i.e. light vs dark colours). In particular the brain derives fine detail and texture from luminance information, but treats chroma as a much more "broad brush" thing. This is why something like red text on a green background is so hard to read: your brain needs to pay attention to fine detail in the chroma rather than the luminance, and its not set up to do that.

(Aside: The technical term for fine detail is "high frequency"; i.e. the brightness or colour is changing a lot in a small amount of space, whereas a large uniform area with only gradual change is "low frequency").

This fact has long been known and exploited for image transmission. The earliest use was hand-tinted photographs before colour film was invented; people found that you could get a surprisingly realistic "colour" picture by brushing coloured dye over a black-and-white photograph. You didn't need to worry about detailed shading, just brush some blue over the dress, flesh tone over the face and hands, green and blue over the background, and hey presto.

Analog colour TV used the same trick. Notice how much less bandwidth was given to the "chroma" (colour) information compared to the "luminance" (i.e. black and white) signal in this diagram of the frequency bands within a PAL TV signal:

PAL TV signal (from Wikimedia)

JPEG and MPEG do this too. The colour information in the image is first "down-sampled", typically to half the resolution in each direction so that one colour pixel is used for each 4 monochrome pixels. Then the monochrome and colour information are separately compressed by a lossy algorithm that typically throws away even more of the colour detail. Your brain doesn't notice. (Incidentally this is why the picture posted by the OP has such a blurry colour grid when you look closely).

So now on to the illusion above. The grid is, in effect, fine colour detail. However the brain ignores the detail of the grid and instead perceives broad swatches of colour over the same areas, leading to the illusion.

  • 2
    This trick is also used for PenTile display matrices. – Ruslan Aug 3 at 10:15
  • For bonus points, even the chroma that's left is usually RGB, which of course doesn't represent the original light at all - it's just a decent match on our visual system. Capturing colour data properly would be ridiculously expensive, and completely pointless, given how little sensitivity to colour we have. After all, colour vision mainly evolved to let us distinguish large blots of colour on a differently coloured background (like red apples among leaves or leopards in the bushes :)). – Luaan Aug 5 at 7:49
  • 1
    @Luaan Excluding rods, normal human vision has three degrees of freedom, so it's possible, given any color, to create a color from RGB that the human eye cannot distinguish from the target color (although depending on the system, you may need to add in black to match the luminence). – Acccumulation Aug 16 at 3:13
  • @Acccumulation Of course, I'm not claiming that this problem is exclusive to RGB - just that the colours are tailored to look "the same" to human vision. All colour spaces have this problem since they're all essentially lossy compression of the original data. It's not like human vision is exactly RGB either - each of the cones has a particular response to a broad spectrum of wavelengths, with different wavelengths producing different strength of response. RGB is just a close enough match that has other desirable features. – Luaan Aug 16 at 7:32

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .