In a comment to this post, a reader asked me to compare the ad hoc color balance adjustment technique that I developed a few days ago with the UniWB technique. I’d not been aware of that approach. It turns out the goal is the same: to get the in-camera histogram to approximate the true raw histogram by adjusting the in-camera white balance. There is an excellent tutorial on UniWB here.
I went through all the UniWB calibration steps with the D4, and got results that were worse than I got with my ad hoc technique. However, I learned two things that I’ve used to make the process I use faster and more accurate.
The first is how to judge the results. Rather than eyeball the histograms in the camera and in Rawdigger as I used to do, I now look at the coefficients that the camera puts in the EXIF metadata to tell the raw processor how to compute the “as-shot” color balance. In order to get the in-camera histogram to approximate the true raw histogram as closely as possible, you want coefficients to be equal. In Nikons, the green coefficient is always one, so you want the red and blue coefficients to be as close to one as possible. Using this technique gives you a numerical measure of how close you are, and tells you which way you have to go to get closer. If the red coefficient is greater than one, add more red. If the blue coefficient is greater than one, add more blue. If both the red and blue coefficients are greater than one, add more green. Maddeningly, the adjustments interact nonlinearly.
Secondly, it’s better to start with a magenta image, tell the camera to white balance to it, and then tweak the resulting in-camera white balance with the fine-tuning controls than it is to set the white balance by color temperature and green/magenta bias. The reason is you get more range that way. With the D4, you can’t add quite enough green through the custom color temperature dialog.
So, here’s my procedure. Open up Photoshop. Create a new image, fill it with magenta on your color calibrated monitor, and make a photograph of it. If you’ve got a Canon, you might as well start with Guillermo Luijk’s values for the Canon 350D. In the sRGB color space, they are: R=162, G=64 B=104. For the D4, using Luijk’s technique, I got R=120, G=64 B=94, but those values didn’t give me very good results. Specifically, they were red-shy. When I used R=162, G=64 B=104, I got pretty close to the right coefficients. Be sure to fill the frame with the color sample. It’s probably a good idea to defocus a little so that you don’t have the detail of your monitor’s pixels in the image, but your camera will average it out later if you don’t bother. Expose for a middle gray. White balance to that.
If the coefficients for the red and blue components of the white balance are within five or ten percent of the coefficient for the green component, you’re done.
If not, go back into the custom white balance dialog and pick “fine tuning” or some such name. Add or subtract red, blue, or green from the white balance. Take a picture of anything. Load the raw file into your computer, and look at the EXIF metadata. If the coefficients for the red and blue components of the white balance are within five or ten percent of the coefficient for the green component, you’re done.
Repeat the last paragraph until you’re satisfied or bored to tears.
Don’t try to get bang on with the coefficient values. As a reader said in a piece of correspondence that I’ll get to in a few days, “We’re playing horseshoes here.” Close is plenty good enough.