Sunday, April 19, 2015

Are there cameras with good skin tones?

A PHOTOGRAPHER ASKS: “What do they mean by a camera with a good skin tone? Is it subjective?”

Well, a camera doesn’t have skin, but some cameras render the color of human skin better than others.

There are five factors that go into healthy skin color:
  • The base connective tissue, which is slightly bluish pink. This contributes to the skin coloring of the very palest individuals, including albinism, where melanins are absent.
  • Blood, which usually gives a red hue due to the capillaries near the skin surface; this can be variable — less in the cold or during fright, and prominent during exercise. Anemia can change blood color, giving pale individuals an ashen appearance. 
  • Pheomelanin is reddish in color and is the main component in the coloring of redheads. It is found in larger quantities in most females, and is found in greater concentration in the lips, nipples, and sex organs.
  • Brown eumelanin — a dark yellow color, actually —leads to blonde hair and yellow and olive skin. When coupled with pheomelanin, it gives brown hair. 
  • Black eumelanin which is colorless and is present in dark skin and in black and grey hair.
  • Eumelanins are typically found in lesser quantities in females than in males.
  • The relative proportions of the melanins changes with age, and may be blotchy, enhanced due to sun exposure. Some diseases may also change skin color beyond the normal range determined by the factors listed above, making nontypical skin color a good indicator of illness.
These combine in various proportions, with melanin predominating where present in significant quantities. Due to these limited number of factors, the range of healthy human skin hue is somewhat narrow, about 30 or 40 degrees of the entire 360 degree HSB color circle and narrower for most of the population. The typical human eye is pretty sensitive to changes in those hues because that is where two of the eye’s cone cell light receptors, the L and M cells, strongly overlap and where color differentiation is particularly sensitive; this strong differentiation of hue is also useful for determining the ripeness of fruit as it changes from green to yellow, orange, or red.

For human skin, with a relatively flat light source spectrum and neutral white balance, in sRGB numbers, Red > Green > Blue, for all ethnicities except for the very lightest and darkest where blue might be equal to or slightly greater than green. With rare exceptions, green is always less than red and never exceeds it. If at the very least, a camera can’t reliably give us those numbers, then the camera can’t give us good skin tone.

If a camera’s automatic white balance algorithm can’t be relied on to work well, then a camera won’t give us good skin color. This process is used to subtract out the color of the light, and some cameras are better than others. Significantly, some camera models will detect faces and use the narrow skin hue range as a means of adjusting white balance.

But for this reason, many photographers will use a manual white balance, taking that one variable out of the equation.

Exposure becomes a significant problem for skin for the very lightest and darkest of individuals. It is easy to overexpose the red channel for pale individuals — turning the skin hue more bluish or greenish — and underexpose the blue and green channels for the darkest individuals, making their skin tone look too saturated red. So if a camera doesn’t give good exposure there will be problems with skin color.

But for this reason, many photographers will keep a close eye on exposure, checking their camera’s three-color histograms. Modern cameras with a high dynamic range can better avoid this over and underexposure.

Most young healthy humans have a skin hue that ranges from a slight bluish-red or rose color through yellowish-orange depending on ethnicity and sun exposure, and these skin hues will vary slightly for each individual, as some parts of the skin are more changed by sun exposure than others, and there is blotching also. So the structure of the red and green channels are critical for getting good skin tones. Certainly those two channels must have a significant overlap, but what is critical is that those channels ought to change strongly in sensitivity over the range of human skin hues — not so much of a change that there is an abrupt cutoff in hue (this used to be a problem in videocameras), but not so little that subtle differences are missed, which is a common problem these days.

Due to the process of metamerism failure, it is entirely possible that a camera won’t be able to distinguish two colors that are distinct to the typical human eye, particularly under some light sources, or, the camera will render differently two surfaces that appear to be identical to the eye. This problem is uncorrectable by the camera. The best that you can do is to select the specific surfaces in your software and force them to be a different color. If your camera has a strong metameric mismatch with skin colors you are in big trouble. A striking and even a bit creepy example of metameric failure is found in cameras that have some ultraviolet sensitivity: melanin in human skin absorbs ultraviolet to protect the tissues below, and so photographs of people with these cameras shows lots of dark blotches from this melanin.

There are a number of Canon and Nikon photographers who prefer using their older cameras for portraiture — generally those made before about mid-2008 — because they say that these deliver better skin color. After that time, these camera makers widened the response of their color channels in order to provide better high ISO performance. But this is a trade-off since subtle distinctions in skin color might be lost. Some of the newer cameras instead deliver blotchy skin hues due to an abrupt change in detected hue while the older cameras detect more intermediate hues. I ought to note that newer models of some other brands have kept good skin color ability.

It is here that good software can help. Converting raw sensor data into a JPEG is a conceptually complex process that is quite error-prone and subject to trade-offs. One of these steps in raw conversion is the color profile which converts the limited spectral data captured by the camera to a standard color space such as sRGB. Designing a good color profile for a camera seems to be more of an art than a science. Some color profiles may shift hues according to lightness, leading to uneven skin color, while a profile which doesn’t do this will have to sacrifice some hues in favor of others in terms of accuracy, which may harm captured skin color. Both Nikon and Canon provide camera profiles specifically designed for portraiture — but better accuracy in skin color is obtained at the expense of less accuracy in other colors.

As modern cameras have a great color depth, we have lots of data to process. We can try to force more overall average color accuracy if needed, but this might end up with more noisy or blotchy colors from some cameras. We can use noise reduction, especially chroma blur to help, and blurring the colors on human skin is more acceptable than blurring luma noise, which gives the plastic skin look. Here a good raw converter, one that uses a significantly greater amount of mathematical precision and advanced algorithms for processing, can help quite a bit. But this precision is a trade-off: I frequently use some good raw processing software, but it can take a minute or more to process an image, something that would be completely unacceptable in a consumer camera which needs fast response time.

Finally, all of us as we age become grayer, as our skin and hair hues become less saturated. Some cameras and software can attempt to counteract this by processing skin hues to be more saturated, and perhaps shifting the hues towards red — which won’t work for all ethnicities. Also, there are some skin conditions that lead to a blotchy look, and so some retouchers will correct for this. Makeup can be problematic if they hues used don’t match up well with the base skin tone — and we are increasingly seeing metameric failure for makeup versus skin color under LED lighting.

So there are many factors that can lead to a camera having good skin tone. Some are under the control of the photographer, others might need special software, and still others are beyond any in-camera correction and so need a good camera, known for good skin color, to begin with.


NewsView said...

Sony cameras, at least the RX-series, to my eye impart a characteristic "Sony" color to JPEG output. Even the warm tones look oddly cool and the blue tones seem more pronounced. The look is not ideal for skin colors, in my view. In terms of non-DSLR JPEGs, I think the color is most pleasing from Fuji cameras and Olympus cameras.

tibor said...

Hi Mark, I have read your post about skintones issues on most sony cmos sensor based cameras and have to agree 100% with your observation : sony cmos sensors all suffer from very low and dull color reproduction, the ossue becomes even worse in mixed light situations.
I purchased my first digital camera about 7 years ago, a humble D40X (10mp CCD) with 2 decent lenses and was actually impressed with the results, even with jpeg files coming out of the box. Then, I started quickly to shoot raw and the Nef files were even more exciting to handle. The files almost matched my 6x7 chrome slides (aside of the resolution) and I printed a few 24x36 inch lightjet shots who came out beautifully with a tiny touch of high pass Filter.

I kept my trusty D40X and purchased a D5100 and later on an "oily" D600 (which was replaced by Nikon for a brand new D610) I also shot with the D300 and D200. While I was impressed with the precision of the D300 AF and the resolution and dynamic range of the D5100 and even better D610, there was clearly something wrong with the color output. I mainly shoot landscapes and portraits, and the colors of sony cmos based cameras are just wrong and dull : they have a tendendy to go magenta and orange, woth greyish brownish shadows. When pulling the shadows, there is a red magenta cast instead of just brightening up the area. And the whole files lack the shiny effect or chrome aspect of my CCD files.

I was an old member of Dpreview and became fed up with some members bashing our observations (many leica shooters also regret the CCD sensor) and ended up understanding that those people did not take the time to compare properly both sensor types using the right software ; capture nx2 may be slow but it offers a way better output then any other software. Some other shooters simply dont seem to bother about color rendition and others just keep upgrading without looking backwards.

I also have a m43 panasonic setup who carries a MOS sensor which is to my understanding a hybrid sensor, and offers quite decent skintones, they still lack some shiny effect on the colors but the color separation and tonalities are way better then on sony cmos based cameras.

The results of my D610 and D5100 files became so frustrating that I began to investigate if there was a way to overcome them without having to struggle with each NEF file in post production. So i took the same nef shots with the same focal lenght on each camera, some landscapes and some portraits, using different lighting conditions. I opened the files in CNX2 and the cmos files were obviously bad looking from the start. I went into the deep settings of CNX2 and created some basic settings that almost matched my CCD files (they still lack some chrome effect but the skintones are now spot on) I am willing to share my settings with anyone who is interested.