What is the difference between auto white balance and custom white balance?

Question

I have used both modes on my Canon 550D and found the results to be quite different, particularly indoors under fluorescent light. Auto white balance seems to produce yellowish images at times, while using the custom white balance seems to produce more neutral colours.

How do these two modes work, and why doesn't the auto white balance do an equivalent job?

I use an image of white objects taken with auto white balance settings in the same lighting conditions to set the custom white balance

P.S. I will try to add images to show the difference

Answer 1

With Auto White Balance most cameras place a greater weight on the brightest parts of the frame. Some newer models, particularly higher end ones, also measure all of the color in each picture and compare it to a stored database. If a match is found the camera applies whatever the database tells it to. The decision is made for each photo based strictly on the content of that one photo.

With Custom White Balance the camera takes your word for it that what is in the middle of the sample picture you provide is what you want to appear as white and applies the same white balance needed to render your sample as white to each photo, whether there is anything the same color in the image or not.

Answer 2

It's easiest to explain what custom white balance does. Custom white balance asks for a particular image that is primarily white or neutral grey. It takes the sample in the image and calculates from it adjustment factors to apply to the response of the red, green and blue sensors such that they are balanced with each other so that equal amounts of each produces a neutral grey. It is a fairly simple calculation and produces high quality results, but the accuracy is dependent on the sample image being neutral grey or non-overexposed white.

Auto white balance on the other hand is when the camera is left to try and guess. It has the advantage of not having to specifically setup a grey card or find something neutral grey or white to take a photo of, however it gives up accuracy since it is only making a guess. The algorithms have certainly improved a lot over the years, but it still can't do things perfectly. Techniques can very from manufacturer to manufacturer and can include such things as looking as the brightest area of the image that isn't overexposed or looking for what appears to be flesh tone. If it can find something it thinks should be a certain color, it is then possible to calculate the correction factors for red, green and blue based on what was sampled vs what was expected.

It is also possible for the camera to look at known profiles for particular types of lighting and see if it matches up with one of these. If it does, then it can have a pretty good idea that it is close to correct. This is why AWB tends to work better in normally lit rooms as opposed to areas where the lighting is distinctly colored (since the custom lighting won't match an established preset.) AWB also tends to suffer in either very low light or very over-exposed images where the contrast is either too low or the image is too bright to be able to effectively sample a range of color and contrast.

As an extreme example, if you took a photo of a blue and yellow card, the camera would have know way to tell if it was bluish light or yellowish light. The card could be white and yellow, blue and yellow or blue and white. Without something else to reference, AWB would fail miserably with no way to guess. On the other hand, custom white balance would have no problem since it was preset from a reference image.

Answer 3

Each light source has a color cast of some sort, daylight is blue, fire and indecent light are yellow, florescent is often green, our brain knows how to "filter out" the color of the light to make it look white.

Auto white balance tries to do the same thing the brain does and automatically filter out the light's color cast, however, the AWB algorithm in the camera is not good enough to compete with the brain (after all, digital photography is fairly new, the brain had million of years of evolution to get things right) - so it often gets thigns wrong.

Manual white balance does not try to guess anything, by looking at a picture of a white (or neutral gray) object the camera can calculate the exact color cast of the light and filter it out perfectly - perfect color every time (if you have the time to set white balance).

(The camera can find the real color based on a wrongly light balanced jpeg because the camera knows what light balance was used for that jpeg)