White balance adjustment: what actually happens?

[milkyman90]

Hi everyone, a question regarding white balance has been plaguing me recently.. it's a simple question, but I sure as hell don't know the answer. I hope someone here does

When you adjust white balance on a RAW file, what does the software (let's say Canon DPP for argument's sake) do behind the scenes? Does it bias the way it reads RGB pixel data or does it shift the data taken from every pixel colour? In other words, what I want to know is whether blue light information is 'thrown away' when you warm up an image. In other other words, if I took a photo which was entirely blue and warmed it up, would it turn warmer or would the blue hue simply disappear?

I guess I could try that last issue myself with a simple experiment, but I'd still be keen to hear from anyone who is privy to the general way in which RAW converters alter white balance.

Any thoughts?

 

[Big Mike]

Good question.

I'd ask THIS GUY.

 

[ann]

well, if mike doesn't know, i surely don't

math is a wonderful thing and that i do know

 

[Garbz]

Short answer.... no short answer and I have guests coming in a few minutes. Be back later

 

[Garbz]

Ok the answer is it would turn warmer, and mess with the hues only slightly as a side effect.

Basically the sensor records 10, 12, or 14bits of linear data. It just records, nothing more. This data as it comes out of the sensor looks like garbage since our eyes don't work on a linear scale. Firstly the data is interpolated so you go from a RGBG beyer pattern on the LCD to each pixel having a RGB value. Then these values are weighted according to white balance. Data is never shifted nor thrown away, it is only ever biased (though sometimes to the extent that it falls outside the displayable range, such as when you crank the exposure all the way up or down). Any settings are also applied, such as saturation, contrast etc.

At least that's how Lightroom does it since it works in a linear colour space. I'm not sure if Canon's DPP works on the same principle. A gamma curve is applied for display purposes, and also during the colour space conversion when you export the files or work in Photoshop.

 

[Dwig]

... In other other words, if I took a photo which was entirely blue and warmed it up, would it turn warmer or would the blue hue simply disappear?...

Actually, if the image was really "entirely blue" you couldn't warm it up. There would be no red, or for that matter green, data to amplify/bias/... in order to balance the image. WB does nothing to create data of its own.

In practice, the common digital sensors are roughly "daylight" balanced. When you adjust WB the data in the are not handled equally during the RAW conversion. The end effect is similar to, though not identical to, increasing the ISO for one color and/or decreasing it for another.

You can see this for yourself by doing a simple comparison. Shoot a picture using very low wattage, or dimmed, incandescent lamps (low Kelvin temp) using a moderately high ISO (something that will produce slight visible noise, but not excessive. Now do two different RAW conversions, one with WB set to 5500K and one with WB adjusted to as low a Kelvin value as your conversion software will allow. If you are using a RAW converter plugin to a photo editor (ACR with PS, ...) simply convert both images to B&W in your editor (not the RAW converter). If you are using a "Super RAW Converter" (Lightroom, Aperture, Nikon Capture NX, ...) save the converted images as color TIFFs or minimally compressed color JPEGs and then reopen them to convert them to B&W.

If you now compare the two B&W images you will see more noise in the image where you made the WB correction. This is because the lamps emitted very little blue and the WB correction process had to amplify the blue data substantially to balance the image. In effect, the blue channel is operating at the equivalent of an ISO three to four stops higher than the red channel to compensate for the weak blue in the low wattage lamps.