Why did your photo come out orange? White balance

Short answer

An incandescent bulb's light is far more orange than daylight; your eye compensates for this without noticing, but the camera only compensates if you tell it what light it's under, or if it guesses correctly. White balance is the camera adjusting its color channels to make a white surface look white again. The automatic setting estimates from the scene's average and can get it wrong; a manual setting made with a gray card is more reliable, but if a room has two different lights, a single setting can't correct both at once.

The color of light: color temperature

What you call white light in everyday life is actually many different colors. An incandescent household bulb, the midday sun, an overcast sky, and the blue light in shade each give off noticeably different mixes of light. The most common way to describe this difference is color temperature, measured in kelvin.

The word "temperature" in its name is confusing: low kelvin values look warm and orange to your eye, while high values look cool and blue. Cinema and photography have used two reference values for years: 3200 kelvin for incandescent studio lamps, and around 5500 kelvin for daylight. Why light comes in these colors is the physics covered by the BI series; here we're looking at what the camera does with this difference.

Your eye doesn't notice this difference to any great extent. In the evening, the page of a book you're reading under an incandescent lamp looks white to you. How the brain makes this correction — color constancy — is the subject of BI-13. The camera sensor, on the other hand, records the light as it is: the same page comes out orange if it isn't corrected.

What does white balance do?

White balance is the camera adjusting the gain of its three color channels so that a neutral surface in the scene produces an equal signal in each. Under incandescent light, the sensor picks up too much red and too little blue; white balance restores the balance by raising blue and lowering red.

Cameras do this in three ways. Presets (incandescent, daylight, cloudy, shade, fluorescent) apply fixed corrections based on the typical color of specific light sources. A kelvin setting lets you enter the value directly. Auto white balance, meanwhile, tries to estimate the color of the light from the image itself.

How does auto white balance make its estimate?

The basic problem with auto white balance is that the camera can't see the color of the light directly; it only sees light that has already reflected off surfaces. An area that looks orange could be a white wall under incandescent light, or it could be an orange wall under daylight. The camera needs an assumption to tell the two apart.

The oldest and best-known assumption is that the average of all the colors in a scene will be neutral gray. In 1980, Gershon Buchsbaum published a model proposing that the visual system estimates the color of the light source from information gathered across the whole visual field, and uses that estimate to recover the true color of surfaces. A simplified version of this idea, known as "gray world," became one of the basic methods behind auto white balance algorithms.

This assumption works in most scenes, because the real world is usually made up of mixed colors. But it gets it wrong when a scene is dominated by a single color. In a frame full of grass, the camera mistakes the green for the color of the light and pushes the image toward purple. At sunset, it "corrects" the orange sky and kills off that warm light. A close-up portrait where skin fills the frame can cause the same mistake: the camera takes the skin's warmth for a deviation coming from the light, cools the image down, and the face comes out grayish and lifeless. This effect can show up no matter what color the skin is or how dark it is, because no skin is neutral. Today's cameras use more advanced methods than gray world, along with scene recognition, but the basic problem is the same: the estimate rests on an assumption.

Manual adjustment with a gray card

The simplest way to avoid relying on a guess is to show the camera a surface you know to be neutral. A gray card, or white balance card, is a neutral surface manufactured to reflect all wavelengths equally.

Here are the steps. Place the card where your subject is standing, facing the light that falls on the subject. Take a frame where the card fills most of the frame, or let the camera's custom white balance menu measure the card. The camera calculates the channel gains that will show the card as neutral and applies that setting to the shots that follow. When the light changes — a cloud passing in front of the sun, or you moving indoors — you repeat the measurement.

If you're shooting in RAW format, white balance can also be adjusted afterward, because the raw sensor data is stored before the channel gains are applied; FS-03 covers this in detail. Even so, shooting one frame of the gray card means you won't have to guess afterward which value is correct.

Mixed lighting

This is where the question on the lesson card comes in: why isn't a single white balance enough in a room with mixed lighting?

Imagine you're filming an interview one evening, in a room with a window. Cool daylight from the window lights one side of the person speaking, and an incandescent lamp on the desk lights the other side. White balance applies the same correction to the entire image. If you set it for daylight, the side lit by the lamp stays orange; if you set it for the lamp, the window side shifts blue. A value in between leaves both sides half wrong. The problem isn't the setting — it's that the scene has two different "whites."

The solution is to equalize the lights. Color-correction gels that bring their colors closer together are fitted in front of light sources: a blue gel to bring an incandescent lamp closer to daylight, an orange gel on the window to bring the light coming through it closer to incandescent. Without a gel, the least damaging option is to turn one source off, or to set white balance for the dominant source lighting the face and leave the other in the background.

You don't always need to correct mixed light, either. The contrast between cool window light and a warm interior lamp, used deliberately, adds depth and atmosphere to a scene. What matters is that it's a choice.

Example — deliberately "wrong" white balance

In film, a color shift is sometimes not a mistake but a storytelling tool. Setting the camera to a value different from what the light calls for warms or cools the entire scene and tells the audience something about place, time, or mood.

Tungsten film is balanced for 3200-kelvin light; used in daylight without a conversion filter, the image shifts blue. On a digital camera, the same effect is achieved by setting white balance to incandescent while shooting in daylight.

This choice came with a cost. The yellow look of the Mexico scenes was repeated in many films and shows in the years that followed and turned into a cliché, and critics wrote that it had become a pattern for making hot, dry countries look dirty and poor, and that the increased contrast also washed out the faces of dark-skinned actors. A deliberate color shift should be set up with thought for whose face it shows, and how.

Common misconception

Summary

1. Light sources have different color temperatures; the eye compensates for this, but the camera only compensates through white balance.
2. Auto white balance estimates the color of the light from the scene's average and gets it wrong in single-color scenes; a manual setting made with a neutral gray card is more reliable.
3. In mixed light, a single white balance can't correct the entire scene; lights are equalized with gels, or one source is chosen. A deliberate color shift, meanwhile, is a storytelling tool, but its effect on skin must be taken into account.

Test yourself

1. (Easy) Which color is a light with a low kelvin value close to?
2. (Easy) Why is a gray card used in white balance?
3. (Medium) Why isn't a single white balance enough in a room with mixed lighting?
4. (Medium) What kinds of scenes does the gray world assumption get wrong? Give an example.
5. (Hard) How is the blue look of Traffic's Ohio scenes explained in terms of white balance, and how would you achieve it with a digital camera today?
6. (Hard) Why must white balance not be measured from a face, and what does this mean for different skin tones?

Answers

1. Orange; incandescent lamps have a low color temperature, while daylight and shade have a higher one ("The color of light: color temperature").
2. Because it's a neutral surface that reflects all wavelengths equally; the camera calculates the channel gains that will show it as neutral, so there's no need to guess ("Manual adjustment with a gray card").
3. White balance applies the same correction to the entire image; if a scene has two differently colored lights, setting for one leaves the areas lit by the other colored. The solution is to equalize the lights with gels or to choose one source ("Mixed lighting").
4. In scenes dominated by a single color; for example, a frame full of grass can turn purplish, a sunset can come out washed out, and a face filling the frame can turn grayish ("How does auto white balance make its estimate?").
5. Soderbergh used tungsten film balanced for 3200 kelvin in daylight without a conversion filter, and the image shifted blue. On a digital camera, setting white balance to incandescent while shooting in daylight gives the same effect ("Example — deliberately 'wrong' white balance").
6. No skin is neutral; the camera mistakes the skin's color for a deviation coming from the light and tries to correct it, so the face comes out grayish. This holds true no matter what color the skin is or how dark it is; that's why measurement is done with a neutral card next to the face ("Common misconception", "How does auto white balance make its estimate?").

Sources

• Buchsbaum, G., "A spatial processor model for object colour perception," Journal of the Franklin Institute 310(1): 1–26, 1980.
• Hope, D., "The Traffic Report With Steven Soderbergh," DGA Magazine 25(5), January 2001.
• Traffic, directed by Steven Soderbergh, 2000.

Next step

BI-13 returns to where this lesson started — the eye's own white balance: it covers Edwin Land's Retinex experiments, how the brain estimates the color of light, and where this color constancy breaks down. FS-03 stays with the camera: it explains the bit depth of a RAW file, in-camera processing, and why white balance can be changed after the shot without any loss.