The Science of Why No One Agrees on the Color of This Dress | Hacker News Reader

The Science of Why No One Agrees on the Color of This Dress

wired.com

56 pointsandrewfong52 comments

Well, without knowing the light source, it's impossible to accurately measure color. They mention applying white balancing to fix it, but this makes an assumption of some characteristic light source:

http://www.cambridgeincolour.com/tutorials/white-balance.htm

Beyond that, there is processing done by the sensor, the photo app, the person who put it online, etc...

That’s not quite it. More precisely, it’s always impossible to measure “color”, because color is a perception in the brain, not a physical fact about the environment.

But it’s certainly possible to make a good guess at the light source by looking at objects in the scene. In this particular case, we can look at the black and white dress at bottom left, which we can reasonably expect will appear neutral to human eyes and most “properly white balanced” camera images.

I’ll agree that the camera used here has a relatively crummy sensor / processing pipeline though.

Yes, but the spectrum of the light hitting the eye is a physical thing.

The physical problem is that multiple environments can make the same spectrum, e.g. is it a bluish light source & white dress or white light and bluish dress. White balancing can fix this if you know the light source, e.g. the sun, but that's either an assumption or something calibrated separately from the image.

That said, the perceptual issues are certainly a problem as well :)

A human can't measure color, but you're measuring something when you count the number of photons coming through a filter.

Looks like someone has rediscovered what Edwin Land pointed out in 1959 [1] which is that the perception of a color depends on what is around it. There was a great video of Land walking with a square of one color from background to background in a single shot and the color of the square appears to change as he changes backgrounds.

[1] http://www.millenuvole.org/f/Fotografia/Per-quali-ragioni-ve...

But the perception is different per individual, not per "what's around it."

Further, the image's colors is changing somewhat spontaneously for some people, including my girlfriend.

Edwin, who went on to create the Land Camera (Polaroid) noted exactly the same thing. And he realized that the reproduction of color was not nearly as straight forward as people thought it should be ('just add the primary colors'). This particular image is a great example because it sits right on the edge of what you're eyes perceive.

Everyone is seeing this image through a color reproducing display. And that display has approximations in terms of the color frequencies it can display. That combines with the way our eyes perceive color and gives us different views of it. On my Moto G for example I can change the color people perceive by changing the brightness of the screen. Same picture just different brightness levels.

Initially I saw white & gold, and I kinda still see it that way. The assumption I noticed I had is that there is a window behind, which gives out a bluer shade of light than the store lights. On closer inspection, the blue tint doesn't quite match that assumption and that the dress must have a blue tint. I cannot unsee the gold shade, but thinking about it, it could very well be the store lights that cast the yellowish tint.

I can see the blue and the white but I cannot see black. Even the darkest images I'm still seeing a gold tint. Can anyone help me out?

The “gold tint” is due to the camera white balance. Does it still look gold if we adjust it so that the background white + black dress is neutral?

http://i.imgur.com/jtjECtR.jpg

I still see gold tint to that image, and it is far from black, which I verified by opening the color in photoshop.

And I'm confused about your point about the camera white balance. It doesn't matter what is causing the "gold tint" only that I am seeing it and others vehemently claim they see no gold tint. The point is there is a perception difference that is not being accounted for. I mean, I understand how the shoddy white balance makes what in reality is black fabric shift gold/orange in a photo, but objectively in the original photograph the bands of fabric are not black or gray but a muddy gold color, so I do not understand how anyone sees it as black.

I still see gold yes :|

Sure. If you look towards the darker parts of the gold, like towards the bottom of the dress, it would be pretty easy to presume that the lightness and gold tinge is due to the lighting / overexposure washing out the image.

Try making the image smaller and squinting at it. Once your brain switches to perceiving the black correctly then the highlights just look like highlights (instead of like gold).

Same thing with me. I have some friends who converted after a few hours though.

I saw this a few hours ago and the dress was very clearly white and gold.

I just looked at it again and it is black and blue.

This is quite the trip.

The only difference I can think of is that earlier I was in a bright, well-lit room, and now I'm sitting in bed in a pitch black room (other than the screen). Maybe my adjusting to the change in my own ambient lighting led to the change in perception?

I saw it on my iPhone, I thought it was clearly gold and white.

Had a discussion about it with a colleague (he saw it as clearly blue and black). I got up to go check his monitor and as I got closer it changed colour right before my eyes.

I went back to my desk and checked my phone and it had changed to blue and black. MIND = BLOWN

I've not been able to see it as gold and white ever since...

I'm so confused lol

It's back to white and gold for me.

I'm also now in a well-lit room next to a large window with natural light as well. I'm curious to see how long this correlation continues.

I used computer vision to prove the image shouldn't be perceived as black and blue to us.

http://austingwalters.com/that-dress-is-gold-and-white/

A bit about myself, I've been doing research into this very thing for the past two years at UIUC.

Your analysis is interesting, but it doesn't make sense to "prove" that an image is a certain color. Color is a construct of the mind. Given that so many people claim to see black and blue, you analysis proves only that your model of vision is insufficient to explain human experience.

I hope you’re joking. Speaking as someone who has spent many hundreds of hours reading color science papers, and another several thousand hours color-correcting photos in CIELAB space, your analysis is absurd.

If you ever come to San Francisco, I’ll be happy to explain why over a coffee.

Perhaps I went a bit eccentric, the same way the news stories were going :)

Still interesting is the differences in who perceives it differently. Take the survey : http://digitalorchard.co/dress/survey.html

I see it as more gold and a blueish white hue. So not taking that survey cause it seems to constrict me into some weird combination I can't agree with.

I see it as lightblue-gold. I can't answer the survey.

EDIT: Perhaps I'm confused because I'm only judging the colour of the pixels on my screen. Funnily, her bag with dots in the LL corner seems "black" to me even though the pixels are of approximately the same colour as those on the dress.

Hm.

I can trick myself into seeing it either way. If I stare at the top part, it's white and gold. If I stare at the lower part towards the LL corner, the brain seems to "shift" the reference point and it's black and blue.

This needs a middle-blue and gold option, 'cause thats what I'm seeing.

Some of those questions are completely unscientific mumbo jumbo.

I haven't seen this link posted in this discussion about this dress.

http://news.yahoo.com/debates-rage-over-color-dress-photogra...

The reason it's interesting is it shows a storefront window with the dress. I think it really helps put the original image into context. Especially check out the contrast between the blue dress and the very white mannequin. Nobody who sees that picture will ever think the dress itself is white.

I’m so confused. Is it people looking at the dress in person that think it looks white?

Someone looking at this photo on a reasonably standard (or reasonably well characterized) LCD display which their eyes have adapted to is going to pretty much see all of the three versions falling between “pale blue” and “dark blue”. Given the yellow background, proper “white balance” would make the dress look even bluer (see example below). Either way, seems like the photo was taken with a pretty bad-quality cellphone camera. Maybe people have just learned to predict color in pictures from some specific cellphone cameras with awful sensors and processing? Or maybe someone spent 20 minutes staring at something bright blue, and then switched quickly to looking at the photo?

What am I missing here? Does anyone here think the image makes the dress look white?

(Also, there’s not much “science” in this wired post at all. Edit: in response to the below comment, the businessinsider post is also pretty shoddy “science”.)

Edit:

One of the nicest online sources about this topic is http://handprint.com/HP/WCL/color4.html#chromainduct ; the rest of that site is in general the best online source about human color vision. (There are better sources, but you’ll need to go to a university library to read them.)

I wouldn’t claim to be a professional color scientist, but I’m about as much an expert on this topic as you’ll ever find among amateurs. If someone wants a 3-hour lecture in person about the science of human color vision, I’d be happy to oblige (but you have to come to an SF coffeeshop for it; could also be a shorter or longer lecture, depending).

Here’s the original image vs. one I “white balanced” to make the obviously white & black dress in the background appear neutral. Does the foreground dress still look white + gold?

orig: http://i.imgur.com/g461GBr.jpg ; white balanced: http://i.imgur.com/jtjECtR.jpg (and I actually cheated a bit toward leaving the blue less intense than I would typically make it if color correcting this image)

In any event, I still can’t explain why people are seeing this dress as white. I now think that it’s likely to be unfamiliarity looking at digital photographs and/or some weird adaptation effect, e.g. looking at a dim phone screen while outside in the shade on a sunny day. Or maybe perception is being colored by a general expectation of white dresses (people often substitute colors in this way when they look at e.g. fire hydrants, human skin, clouds, or grass, interpreting the color based on past experience despite weird lighting conditions).

Yes, to me the dress looks gold and white. Granted, we're looking at a side of the dress which is in shade (thus there's a tint). But, in my mind, the dress is clearly white and gold.

My wife, on the other hand, first saw it as blue and black. Later it seemed to change for her.

The only way you’d get a “tint” like this on a white dress in the shade, indoors, in a room with yellowish lighting (look at the background) is if you shone a bright blue spotlight at it.

In real life, you are right. However, this picture was taken with a camera, which did some auto white balancing internally. Perhaps the camera chose to use the yellow background as the reference for white-balancing, causing the less "white" dress a bluish tint.

You cannot just consider the natural image here. I am sure that if we got everyone to examine the dress in real life, there would be a lot less confusion.

I bet a good part of the confusion is caused by faulty in-camera white balancing.

That's strange, the 20-or-so people I've read that had it changed all had it change in the other direction--white & gold first, then blue & black later.

FWIW I read that the real dress, as observed in person, is blue & black.

EDIT: Girlfriend reverted. If nothing else, this is good for a giggle.

Here's the real dress, from the supplier's image of it:

http://s3-ec.buzzfed.com/static/2015-02/26/22/enhanced/webdr...

I think this is supposed to be another picture of the actual, specific dress in question: http://media.tumblr.com/ec387ec0bb03230268a9e905d74097d9/tum...

In my talking about it on FB and with my girlfriend, yes people believe very much that it is white and gold.

I have only seen blue & black.

My girlfriend had it change color after about an hour, and started seeing blue & black. The lighting in the house & brightness on her screen were unchanged.

I didn't think "the science behind" it was very scientific. I read a better article, with a (to a layperson) seemingly plausible explanation regarding rods & cones & different types of light combinations: http://www.businessinsider.com/white-and-gold-black-and-blue...

I was staring good 15min to this picture and could not see anything white about it, so I'm surprised too. Another reminder that we cannot trust our perception even on such "obvious" cases.

People are so used to seeing pictures from phones now, and many phone cameras either have shitty auto white-balancing or the user sets it up once for tungsten or outdoor light and then forgets to change it, so everything is shifted blue or orange, respectively.

The only thing this scientifically proves is that marketing agencies are getting better and better at viral marketing.

Imagine that, that very same dress is now back in stock. Purchase one today! http://www.romanoriginals.co.uk/

It looks like a white and gold coloured dress, which was accidentally washed with a blue sock :)

Slightly tangential, are there any computer vision algorithms that react to optical illusions in the same way human vision does?

You’ll need to elaborate. There are hundreds of different “optical illusions”, because human vision is extremely complex with numerous processing steps and many types of adaptation. There’s no single “algorithm” that responds to images the way a human responds.

If you can narrow your question down to a few effects, and figure out what you want your algorithm to do (classify images? print a nice picture to hang on your wall compensating for different scene lighting? detect object boundaries? predict what color name a human will attach to a particular object?), then it should be possible to find or make a program that does that. (At least in principle. Really depends how sophisticated your goals are.)

I don't want an algorithm to do anything. I am not in the field, and I was curious whether any computer vision systems account for any (not all) optical illusions that humans are susceptible to. It's just a curiosity.

What I haven't seen mentioned as a factor is eye color.

I have blue eyes and see the dress as blue and black.

The pigment in your iris has nothing to do with the colors you see.

Eye color affects contrast and glare perception.

There's a joke on film sets: Why don't more Directors of Photography smoke? Because it takes 3 hours to light.

Typically the director chooses one or more angles for viewing the action of a scene, and then the production designer makes sure that whatever appears within the frame (wallpaper, props, furniture) looks right. Then the Director of Photography's, or DP's, job begins.

The basics of lighting involve setting up areas of light and shadow, of the kind you know if you enjoy black-and-white films. But from there, building a consistent color balance for a scene is an enormous amount of work. There are multiple different sorts of lights (incandescent, flourescent, LED, HMI), each with different color temperatures (often switched in and out within the same fitting), which are in turn modified with varying thicknesses of colored gel and which have to match the diegetic light sources - ie those which will appear on screen as part of the 'story world'. Sunlight is blue. So's moonlight, but not as blue - if it's too blue then it looks like a blue light instead of a white moon. Interior lamps have their own color temperature, which usually tilts orange. Then you have to deal with multiple lights; typically there's a key light illuminating the subject, a back light which helps to isolate the subject from the background (so the actors don't look like they have a tree growing out of their head or something - longer lenses flatten perspective and make it hard to differentiate foreground and background at narrow apertures) and a diffused fill light to balance these more intense point sources and obscure their artificiality, without losing the selective quality of the illumination. Each of these has to be color-balanced separately. Complex scenes sometimes require 6 or seven light sources. If you're shooting interior scenes and it's not essential to look out the window, it's often preferable to just black out all the windows or shoot at night, and employ artificial lights instead of the sun.

Here's an example of what it takes to make a scene look 'natural': https://plus.google.com/photos/109568505224923852682/albums/... What you can't see here is that the lower halves of the windows are covered with multiple sheets of neutral-density filter gets, gut to size (whereas the unfiltered upper parts of the window are just a glow of overexposure). that's several hours of work before it's even worth turning the camera on.

Shooting outside in sunlight should be simpler, right? Nope - you need reflectors (White? Gold? Silver? it depends), flags to take light off the hot spots, large silks for diffusion (which require extensive rigging to hold what is essentially a small parachute and prevent it flapping in the breeze), and people able to reliably adjust everything during the shot if the camera or the subject is moving. Also, the motion of the earth around the sub means that the position of everything needs to be tweaked about every 15 minutes as you do more takes and shoot things from different angles, all of which have to look completely consistent later on even if the source footage was shot hours apart. Also the color of the sun changes depending on the time of day.

Lighting for film is a massive headache most of the time* which is why you just don't see certain clothing combinations on film. A dress like this is enough to make an industry veteran cry.

* unless it's mild and overcast. Moderate cloud cover basically doubles productivity for daylight photography.

To me it is clearly blue and gold/brown. So what does THAT mean?

It looks blue and black to me. Can't see any white.

Very interesting analysis.

Today's xkcd: http://xkcd.com/1492/

I don't think this explains it fully. This shows how the perceived color has a dependence on the background, but not how it has a dependence on the observer as well.

In fact, for me, I saw it as white and gold at first. Now, several hours later, it's black and blue and I can't see the gold at all. The background of the image didn't change.

On the dress, I see blue & black.

I see neither white nor black in that comic.

I do see gold, and I see a lighter blue on the leftmost dress.

Anyway, I see his point (there's a bit of an illusion), but I don't think he's mirrored what happened with the dress. The dress' background is not changing, and people view it differently.

Reply on news.ycombinator.com