Lighting, White Balance, and Skin Tones

[jcs]

Part of the challenge to getting good skin tones is lighting. Many lights have a green or magenta bias (cheap LEDs, fluorescents, CFLs, etc. (some better than others)). This can be especially troublesome when the lights are far away from full spectrum. You can think of light as a lookup table multiplier and real-world objects having coefficients which determine what color they reflect. For example a fire truck has mostly red with less green and even less blue. If you shine a blue light at a fire truck then try to correct it in post, it won't look very good. If you were to shine a blue laser at a mostly red object, you'd see mostly black. Here are some example spectra:

Now it should be clear why shooting in daylight or incandescent we get the best results. The color multipliers are more uniform across the spectrum, and all the subtle off-shades are captured. The worst example is fluorescent (some fluorescents are much better than this example, especially those designed for use in photography/interior design).

White balance (WB) is most simply applied to an image as an RGB multiplier, for example (1,1,1) (white, won't change the image, (1,.8,.6), a WB to warm up an image, etc.). It's clear that if we process the WB this way white may look white, but other colors will be sorely misrepresented due to spikes in the spectra (the next more complex step would be to use an offset in addition to the multiplier). Some colors will be way too saturated, some the opposite. Do modern camera systems use (3D?) lookup tables or complex algorithms to try to compensate? I would expect that presets could do this, and by looking at the color temp they could guess at the type of light source (low K = the incandescent curve, high K the daylight curve, then interpolate for in-between values).  Halogen has a yellow peak, cool white LED has a blue spike, and warm white has a green peak, etc. Ideally we'd want the camera system to compensate for these cases too. It's likely the cameras have algorithms for figuring out the spectra of the light from histograms and other means, where some cameras to a better job than others. Canon has paid particular attention to skin tones. This attention could be all the way down to the sensor+hardware behavior, which would put less of a demand on the software side to correct (lower power as well). 5D3 RAW appears to show this.

One of the reasons computer generated imagery can look fake is the simple light model (just R,G,B multipliers for lights and object materials). Realism is improved when taking into account spectral properties: http://en.wikipedia.org/wiki/Spectral_rendering

In the early days of computer graphics, we had to write our own renderers from scratch, the most basic models using simple scalar coefficients for both lights and materials. Even with fancy texture maps, specular modeling, filtering, and post processing, the CGI had a fake quality to it (even all they way up to Pixar's Renderman used in many movies). Now we have renderers which take the spectral characteristics of light into account, where the final results are looking much more realistic (some would argue photorealistic): 

http://www.maxwellrender.com/, http://en.wikipedia.org/wiki/Metropolis_light_transport

So when our real-world images start looking fake or plastic, the simplest first step would be to take a look at the lighting. If we set white balance so white is white, skin tones might be off. So we adjust skin tones to look more correct, now the background and even whites might be off. That's a good example of low quality lighting and/or weak color science in the camera. One reason RAW cameras can look so amazing is that much more powerful color science can be used in post (too complex to run on the camera).

Currently the easiest way to deal with lighting issues is to shoot a color chart, then using Resolve's Color Match feature. This can work remarkably well in getting two cameras to reasonably well match. It's more work to shoot the charts, and it doesn't always result in a pleasing look, especially for skin tones. Kudos to the camera makers with good skin tones under most lighting conditions!