What color was the dress?

Short answer

The dress's true colors were dark blue and black, but the photo left out one clue: what light was falling on the dress. As the brain calculates the color it sees, it estimates the light and subtracts it from the calculation; because the photo left that estimate ambiguous, some people assumed a bluish shadow and others a yellowish lamp, and the same pixels turned into two different dresses.

One photo, two camps

On February 26, 2015, an ordinary photo of a dress, taken in a store before a wedding, was posted on a social media site, and within a few hours the same question was being asked all over the world: was this dress white-gold or blue-black? What set the argument apart from an ordinary disagreement was that both sides were equally certain. Two people looking at the same screen side by side could not convince each other of what they were seeing.

There was no ambiguity about the dress itself: the item in the manufacturer's catalog was a royal blue dress with black lace. The ambiguity was in the photo. Within a few months, researchers turned this spontaneous experiment into a systematic survey and showed that the split was genuinely large.

The brain estimates the light

The light that reaches your eye is the product of two things: how the object selects light, and the light falling on it. The brain only sees the result and has to decode the object's color from it. To do this, it constantly makes an estimate: "What light is illuminating this scene?" It subtracts the color of the light it estimates from the calculation and counts what remains as the object's color. This illumination estimate is usually so fast and accurate that you never notice it happening. That's why you see a white sheet of paper as white whether it's under the noon sun or an evening lamp. This correction is called color constancy, and how it works is explained in detail in BI-13.

Most scenes offer plenty of clues for this estimate: shadows, glare, familiar objects nearby. In the dress photo, these clues were weak. The image was overexposed, the background was bright and blurred, and the dress filled almost the entire frame. The brain had to decide which light to subtract, and there were two reasonable answers.

First answer: the dress is standing in a bluish shadow. Shadows are tinted blue because they're lit by skylight. If the brain subtracts that blue, the blue pixels turn white and the brownish pixels turn gold. Second answer: the dress is under a yellowish artificial light. If the brain subtracts the yellow, the same pixels stay blue and black. Both sides were doing the same calculation; what differed was which light they fed into it. Lafer-Sousa and her colleagues interpreted their results exactly this way: each group was subtracting a different illumination assumption from the calculation.

Morning people and night owls

So why did some people assume a shadow and others a lamp? One study looked for the answer in people's daily light habits. The hypothesis was this: someone who spends most of the day in daylight might be more inclined to assume daylight and shadow in an ambiguous scene; someone who spends more evening time under artificial light might be more inclined to assume a yellowish lamp.

This result needs to be read carefully. The study shows a correlation, not a definite cause-and-effect chain between people's sleep schedules and how they see the dress. Participants reported their own habits, and the effect wasn't in the same direction for everyone — it showed up on average. But it strongly supports the broader idea: faced with an ambiguous image, the brain brings in an assumption drawn from past experience.

Example — skin tone on a video call

Imagine you're on a video call in the evening, next to a lamp with yellow light. If you were in the same room as the other person, you would see their skin in its natural color, because your brain would estimate the room's yellow light and correct for it. On screen, though, the face often skews orange. The camera tries to make a correction too, but it can't read the scene as well as you can. On top of that, the image on screen is now being judged together with the light in your own room; your brain has no way of knowing the light in the other room.

This happens the same way for every skin tone. Light or dark, the light reflected from any skin carries the trace of the light falling on it. When the correction fails, light skin tones skew toward pink or orange, dark skin tones skew toward reddish-brown; and if the skin is very dark, the camera's exposure adds to the problem and detail can be lost. How cameras set white balance is covered in FS-01, and how to capture different skin tones accurately is covered in FS-11.

Common misconception

Summary

1. The dress was actually royal blue and black; the argument arose because the photo left its illumination clues ambiguous.
2. As the brain calculates the color it sees, it estimates the scene's light and subtracts it; different light assumptions turn the same pixels into different colors.
3. Everyday light habits can influence this assumption; studies show a correlation, not a definite cause.

Test yourself

1. (Easy) What were the dress's true colors?
2. (Easy) In the survey of 1,401 people, what was the most common answer?
3. (Medium) Why can the same image evoke a different color in two people?
4. (Medium) If the brain assumes the dress is standing in a bluish shadow, which colors is it inclined to see?
5. (Hard) Does morning people seeing the dress as white-gold more often prove that sleep schedule directly changes color perception? Justify your answer.
6. (Hard) Why does a face that looks natural to you in the same room skew orange on a video call?

Answers

1. Royal blue with black lace ("One photo, two camps").
2. Blue-black; 57 percent of participants gave this answer ("One photo, two camps").
3. As the brain calculates color, it estimates the scene's light and subtracts it. When the photo leaves the illumination clues ambiguous, people assume different lights and turn the same pixels into different colors ("The brain estimates the light").
4. White and gold; once the blue of the shadow is subtracted, the blue pixels turn white and the dark pixels turn gold ("The brain estimates the light").
5. No. The study shows a correlation; participants reported their own habits, and the effect showed up on average. The result supports the role of an experience-based light assumption but doesn't establish a definite cause-and-effect relationship ("Morning people and night owls").
6. In the room, your brain estimates the yellow light and corrects for it. On screen, the camera's correction can fall short, and you judge the image together with the light in your own room; you have no way of knowing the light in the other room ("Example — skin tone on a video call").

Sources

• Lafer-Sousa, R., Hermann, K. L., and Conway, B. R., "Striking individual differences in color perception uncovered by 'the dress' photograph," Current Biology 25(13): R545–R546, 2015 (PMC4921196).
• Wallisch, P., "Illumination assumptions account for individual differences in the perceptual interpretation of a profoundly ambiguous stimulus in the color domain: 'The dress,'" Journal of Vision 17(4):5, 2017.
• Brainard, D. H., and Hurlbert, A. C., "Colour vision: understanding #TheDress," Current Biology 25(13): R551–R554, 2015.

Next step

BI-13 unpacks the mechanism behind this lesson's correction: it shows how Land's Retinex experiments reveal the way the brain estimates illumination, and the limits of color constancy; the same gray square looking yellow and blue in different surroundings is the laboratory version of the dress. FS-01 moves on to how the camera does the same job: it covers the estimation logic of auto white balance, the gray card, and mixed-light rooms.