http://www.wired.com/2015/02/science-one-agrees-color-dress/

As a visual neuroscientist I think a lot about how we see the world around us. And so I’ve found the scientific and celebrity controversy around #thedress to be especially fun and exciting. Most of the scientific pundits have concluded that the dress is black-and-blue, and they have offered up an illusions-in-the-brain explanation of why some people see the dress instead as white-and-gold. Yet after thinking thoroughly about this photo, looking at it on a number of different screens, and speaking with some of my lab partners, I’d like to offer my point of view. Yes, there is an illusion at play here that affects our brains, but no, it is not that illusion that causes it to look differently to different people: that difference is caused by a mundane photographic effect.

First, the illusory explanation (it’s black-and-blue but only appears white-and-gold) arises from what we scientists call “color constancy.” It’s the process by which we can recognize the same object under different light sources. My favorite example of this effect is from Dale Purves’s lab, and is shown in the image of the Rubik’s Cube. You see the brown central chip on the top and the orange central chip on the front-face? In fact they are identical in color and only different because of their context: the brain does a computation of the lighting on each surface and calculates the surface properties of each chip based on those lighting properties. This illusory process helps you determine the color of things irrespective of the light source. So I agree that my colleagues are not wrong that the brain can do amazing tricks in your brain that make objects look surprisingly different under different lighting conditions.

Courtesy of Beau Lotto and Dale Purves

But that doesn’t mean that’s what’s happening with the dress. It’s a poor image to work from, but perhaps if we analyze the photo more closely with an eye to determine the nature of the lighting sources, it will inform our decision.

The best analysis of the image I’ve seen so far was done by photo editor Neil Harris, in Roger Adams’s post at WIRED.COM. The WIRED team did a Photoshop analysis in which they analyzed the colors from different parts of the image and, most importantly, varied the white balance of the image. They showed that the dress looks different depending on the white balance. You can also prove this to yourself by simply squinting while viewing the original image on Tumblr… it turns from a white-and-gold image to black-and-blue.

http://www.wired.com/2015/02/science-one-agrees-color-dress/

So this explains why different people see it differently: they are looking at it on different screens set to different brightnesses (or white balances). It has nothing to do with your brain—it’s just an optical issue of light transmission. This also means that we may not need to invoke an illusory process in the brain at all, and that we could potentially identify the nature of the dress through a photographic analysis.

Note that changing the white balance is not changing the colors. Why would black lace look golden, and not green or red, when you crank up the luminosity? Shouldn’t it be gray? Given all the speculating everyone’s been doing, I’m surprised that nobody’s mentioned specularities. These are the mirror-like reflections on the shiniest part of any the object. They provide the best clue of the light source shining on an object, because they are literally reflecting the purest available sample of photons that came from the source. So by analyzing the specularities, we can see what color the lighting is. And when we see what color the lighting is, we can then determine what color the cloth of the dress might be. This matters when you think about the brain because, as you saw with Purves and Lotto’s Rubik’s cube, how we see light dictates how we interpret the world around us, whether it’s outside, inside a store, or on your smart phone screen. It’s an imperfect analysis of course, because the specularities are not perfect mirrors—they are also affected by the cloth’s color itself—but let’s take a look and see what we can determine.

In this case, let’s look at the jacket part of the dress in the top right of the picture. See how shiny it is? That’s where the specularities are the strongest, and there we therefore have the strongest clues to the source of light shining on an object. If you keep your eye on the shiny part of the jacket, you’ll see that its white color changes when the white saturation of the photo changes, but not otherwise.

It would be best if we could analyze specularities made from sunlight. The light that comes from direct sunlight, at midday, is more-or-less exactly known. So if we knew that the dress was bathed in sunlight, we could analyze the specularities. This would help us determine the true color of the cloth, account for different ways that different cameras might operate, or different settings on our display screens, etc. But the original photo wasn’t taken under sunlight, it was taken inside a mall, presumably by a smartphone camera. The camera in the phone likely corrected for color and brightness on its own, reacting to the bright light shining in the background, not the dress in the foreground. Yet even with the bright background, and the color correction, the specularities don’t change.

So what do we see here? Well, the specularities in the original image are bluish. This suggests the light source is either blue, or the cloth is absorbing the non-blue photons (from the lights shining in the mall), and only reflecting the blue photons. Since they aren’t, it’s possible that we are seeing a white-and-gold dress under a blue light source. So how do we know which effect is at play? Is it an illusion or is it photographic? Well, we know that the white balance effect is at play: the WIRED analysis makes that clear. So it’s photographic at least in part. But is there an illusion at play too?

I think there is. This morning I interviewed with Brenna Ehrich at MTV.com and told her I thought that the difference between viewers must be photographic and not illusory. And that’s true. But in investigating this issue, my assistant Max Dorfman saw something interesting in the magnifying tool on the dress manufacturer’s website. The white transparent rectangle that magnifies the dress made it look white and gold, despite there being no issue of light sources. This means that the illusion of gold lace can happen with black fabric when it is surrounded by blue fabric, when a white transparent film is placed in front. From a neuroscience perspective, this means our brain is indeed applying a golden color where none exists, as a color constancy effect. So that’s it then: there is both a light balance effect (as WIRED’s demo shows), to explain why different people see different things on their different screens, and an illusory color constancy effect that contributes to why we see gold instead of gray when its surrounded by blue.

http://www.romanoriginals.co.uk/page/home

Either way, I don’t think that the reason people see the dress differently from each other is an interesting brain process. Rather, it is a mundane differences in how people have viewed the image on their electronic display screens (phones, tablets, laptops, etc). So now we know that Taylor Swift and Ellen Degeneres set their phone screens to different brightness levels than Justin Bieber and Julianne Moore. You’re welcome.

 

Update: Several readers have commented that my interpretation of the dress can’t be right because more than one person can see the same image differently at the same time--and on the same screen. I was initially puzzled by this because I do address this issue above… but on second reading I see that I was too blithe in my discussion individual differences in perception and that has led to confusion. So I here’s an update to clarify. I made four points: 1) the change in color of the dress is an illusion related to what scientists call “color constancy”; 2) that photographic effects are at play in this specific image as well; 3) that the difference between people can be a function of criterion (this is the belief, upon looking at the dress, of what the observer is looking at—think of this as “you see what you want to see”--criterion also includes your prior experience with seeing the world, which also affects how you see); and 4) the role of viewing conditions in why people see things differently.

1) Color constancy is described above—the brain’s ability to interpret the surface color reflectance of an object as unchanging under different sources of light having different wavelength distributions—but what should be made more clear now is that the dress is really more of an example of a failure of color constancy. That is, if color constancy worked well for everybody all the time, we would all see a black and blue dress since that is what is physically being displayed. The issue (as best described in Pascal Wallisch’s terrific post on Slate), is that the brain must guess at the surface properties as a function of what it believes the color of the light to be.

In the washed out lighting conditions, the dress appears white and gold instead of blue and black because our color constancy systems take the black lace, which is grayed out due to the lighting conditions, and fills it in with gold. This happens because the grayed-out royal blue cloth from the dress (now light blue in these lighting conditions) can be interpreted to be white with a blue light source over it, and since gold plus light blue is gray, the brain assumes the gray lace is also bathed in blue light and therefore must be gold. Another way to see this, however, is that it’s a blue and black dress seen through white glare. So some people who assume the dress is blue and black will see it that way instead of white and gold.

2) There are weird photographic effects. In my discussion of the WIRED.COM photographic analysis above, you can see that in the high white balance image there is no blue on the white cloth, yet the lace is nevertheless golden; the golden color here cannot be due to color contrast effects because there is no blue to make the gray color look gold. Indeed, photoshop analysis reveals that the pixels truly are gold. This indicates that the blue-and–black dress may be bathed in dim gold light, and that if you crank up the white balance to see the dress more clearly (thereby washing out the color contrast), you are left with white and gold. These photographic non-linearities also contribute to the confusion and difficulty in maintaining proper color constancy of what is actually blue-and-black dress.

3) This issue of why people see the dress differently is the heart of the debate, moreso than the issue of how a blue-and-black the dress appears to be white-and-gold (which is a well-understood illusion, at least to visual neuroscientists). OK, there’s an illusion, but why are people seeing it in different ways? Here I stated that our criterion going into the process was certainly at play. If you are a Taylor Swift fan you may be influenced by her. Or if you were charmed by Julianne Moore’s performance in Still Alice you may instead be influenced by her view of the dress. Or you may have a belief that the lights source is blue (perhaps you are looking at the dress from outside during the afternoon with a clear blue sky behind you), or instead gold (because you are standing in an enclosed mall with artificial lighting). Whatever your prior beliefs, these are called "criterion", and they have an important and powerful effect on how you interpret ambiguous data in perception. It is this point that I failed to emphasize in my original statements above, though they most likely explain how two people looking at the same exact image of the dress on the same device at the same time will have two different interpretations. The alternative to this possibility is that there are two completely different populations of humans intermingled on the Earth: one with a propensity for black-and-blue and one for white-and-gold. But that is too silly to take seriously.

4) Different viewing conditions is another source of difference that I proposed, and which I emphasized because I viewed it as a possibility that other pundits had missed and I thought it should be discussed. Namely, if you look at the WIRED.COM analysis you see very different dresses depending on the white balance (or if you squint your eyes). So one possible accounting for the different things people see on their different phones or tablets is that the settings and viewing conditions are different.

I hope this clears up that there are several possible sources of ambiguity in this illusion, all of which could individual explain the effects.