By Carl Weese
The first part of this series described what constitutes an accurate white balance, and how to get there with a digital camera. Part II will deal more with visual judgment and the idea of "convincing color." Here are a pair of files, a test and a picture, shot under winter overcast morning light, viewed together in an ACR filmstrip window:
There's a big white balance error here (the overexposure and even the burned out red channel, which is actually caused in part by the white balance error, don't matter because the gray area of the card has plenty of tone). A number of digital cameras I've used get confused by a subject with a lot of green, especially under cool lighting. At risk of anthropomorphizing, it seems like the AWB software says "Wow! look at all that green, we must be in fluorescent light, better crank up the magenta on the Tint slider." This is usually accompanied by a Temp setting that is lower than normal daylight, when it ought to be higher. The WB here is 4800/+19. That's actually a setting you might find correct for a mix of fluorescent light with cool daylight. Shooting on a rainy day in an evergreen forest I've seen AWB go for color that reads out 4100/+35, which actually is about right for warm-white fluorescent tubes.
The "real picture" was shot on AWB as well, but the software has done much better. The numbers are 5000/+1. The grass and the winter forested hillside in the background strike me as quite convincing. The AWB had much less green subject area to deal with and the slice of overcast sky gave it something to grab onto. It "feels" about right for the cold winter light, but there's a problem. What drew me to the subject was the stone of the monument, which has very unusual colors, at least for around here. So let's select both files, click the eyedropper on the WhiBal card and then switch the main view to this picture again.
This nails the unusual colors of the monument, though it has lost the sense of cold winter overcast. Still, for this picture, I'll take accurate presentation of the stone over a more convincing rendition of the light. The numbers are 5500/-7, which is in the expected range for overcast light recorded with this camera.
One of the first things you should do to begin using WB calibration is to shoot multiple tests with your equipment in standard situations like clear sunlight, and overcast light. Be sure to do these tests out in the open where there is no reflected light from bright colored objects (including your clothes!) to contaminate the reading. While in theory sunlight should be 5000/0, the raw development software is seldom that perfectly calibrated to the camera. The Pentax K20D used here gives neutral color under clear sun at 5300/-6 in ACR, while an Olympus E-1 I've used for years comes in at 5400/+14. It can be interesting to know how far from standard the color of the light is in a given situation, and for that you have to know what your camera's "standard" numbers are.
Another pair of pictures, this time shot in pretty weird conditions. It was a dark day, light rain falling, and the sidewalk and street were heavily overhung by a green canopy of leaves—there's a row of trees growing at the top of the retaining wall. AWB has made a valiant effort to cope, turning in numbers of 6750/+2. AWB has compensated +1450 temp and +8 tint compared to where it would set for ordinary sunlight, because it knows this sure ain't clear sunlight. Actually, it really felt that cold and blue standing there in the rain, but these colors are pretty livid and I know that they'll look even stranger printed than they do on the monitor.
So here's the test frame after clicking the card with the eyedropper. Look at those numbers, 8800/+4. That's very weird light indeed. The card and the hand look good. However...
This is all wrong. The test card reading has been applied to the picture file. Okay, those are accurate renditions of the subject colors, but it totally misrepresents the scene. Everything seems bathed in a glow of clean sunshine. The color of the light (sunny) is totally at odds with the tonal quality of the light (heavy, soft overcast). The whole point of the shot is lost. How to get back the feel of the day and the strange location?
Some time back I was working on a set of pictures made in overcast light and through tedious trial and error finally arrived at white balance settings that seemed convincing to me. Some of the takes included WhiBal test shots so I reverse-engineered things a bit. I applied the adjusted Temp/Tint settings from the good-looking files to the test card files, then used the eyedropper to measure the r/g/b readings on the test card. It turned out that the white balance that gave convincing portrayals of overcast light had r/g/b values where the g was a couple points higher than the r, and the b was about five points higher than the g. Then I went looking through my archive folders of raw files for pictures that, after laboring over, did a good job of presenting really dark, stormy light—pouring down rain or threatening snow. Repeating as above, I found that for these g ran about five points over r, and b about ten points over g, instead of the dead-even numbers of an accurate gray card calibration.
Right after I'd finished doing this, Oren Grad asked me to look at some snow scenes under overcast skies he'd struggled over to get the color just right before putting them up on his website. I thought they looked perfect on screen, so I pulled them down to my desktop and tossed them into Photoshop so I could get the eyedropper on them. The light gray snow showed r/g/b values such that g was a couple points higher than r, and b was about 5 points higher than g. To have a colleague confirm your results by request is nice, but to find him doing it by independent discovery means you might actually be on to something.
So, I selected the test file again from this screen, then hit "Select All," then clicked the Temperature slider to activate it, and placed the cursor on the test card. Next I used the keyboard down arrow to tweak the temp setting lower, watching the numerical rgb readout of the card and the thumbnail of the other file. When the temp setting got down to 7250, the color of the thumbnail was starting to look pretty good and, sure enough, the r/g/b numbers for the test card were 174/179/188. I switched the main view to the picture file and got this:
Much more like it. So here are three white balance versions, one set by the camera's software, one set as accurately as possible using the test card, and one set to what strikes me as a convincing interpretation of the scene as I experienced it. None of them is "correct." You don't have to agree with my choice. You might think the scene works best at another intermediate balance, or even farther to either end of the scale.
In Part I, I mentioned that portraits probably ought to have accurate color, but I'll modify that to say that I suspect a commercial portrait photographer won't get any complaints if she routinely tests for accurate white balance and then bumps the temperature up 300 or 500 degrees. Warm is happy and friendly.
Household tungsten lamps have really low color temperature. Even though our eyes adapt to this so that we don't generally perceive household lighting as yellow or amber, pictures in living room lighting often don't look right when the color is fully corrected for accuracy. It's the other side of the coin from my rainy day under the trees picture. Maybe this is because somewhere in the back of our heads we think, or feel emotionally, that indoor light should be warm. Paintings from hundreds of years before the invention of color photography often depict fireside and candlelight scenes with a really warm amber palette far removed from an accurate presentation of subject colors. As with the previous example, a test exposure that tells you what white balance is accurate is a helpful reference even if you decide that a warmer rendition will make a more convincing presentation.
If you are in light that is likely to fool the camera's AWB and you plan to use the camera's histogram for exposure evaluation, it's a good idea to select a manual balance or attempt a custom white balance to get the color closer. Even though you can correct the raw file for white balance based on your exposure of a test target, the camera histogram, based on a badly color-balanced camera JPEG, may be so far off that color correction will end up clipping one or more channels. If the WB is set close to accurate color, the histogram will be a much more accurate exposure indicator, and you can still fine-tune the raw files for color using the test card exposures. Note, changing the white balance setting in-camera doesn't have any effect on the raw data, but making the JPEG-based histogram more honest can avoid mistakes based on a faulty histogram.
You can also save specific white balance settings that you find recurring in your work, like those clear sunlight numbers I mentioned earlier. With PS, you can save a white balance setting in ACR as a "custom subset." You can then apply it to new files either in ACR or Bridge. It can be informative to toggle between what the camera's auto white balance came up with and your standard daylight or overcast settings, for example. If you do a shoot in a place with weird light, you can save a custom setting for that, and it will prove useful during your editing. You can always delete the setting to keep the panel less cluttered, after you're done with that shoot.
White balance is both a technical tool and an aide to expressive printing. Use of a neutral test target makes quick work of getting the most accurate color your camera can come up with in a given situation. At the same time an accurately corrected test file gives you a helpful visual reference point for expressive use of color that gets across your personal interpretation of a scene.
See more examples at Working Pictures