Skin Tone color corrections, CYMK or RGB?

[hpycmpr]

Originally Posted by jayk2

Almost every tutorial and book says to edit skin tones using cymk and basically make sure the yellow reads 3-5% higher then Magenta. I work in Adobe RGB and wondering if that percentage rule applies only to CYMK? or can I apply that rule when using RGB?

With the exception of books by Eismann and Margulis, most of the books tend to teach correcting with rgb, skin tones or otherwise.

I almost use only rgb exclusively. One reason is my weakness in knowing how to mix channels, and another is the fear of losing data when changing modes between rgb and cmyk. Margulis goes into quite a bit of detail on channel mixing. I can follow his theory, but find it hard to put into practice without detailed tutorials.

[hpycmpr]

Originally Posted by Duv

I myself haven't a clue how to adjust skin tones using RGB. I've always converted to CMYK to do my skin correction and then back to RGB. There seems to be a lot more information on CMYK skin correction. Although I haven't noticed anything, I'm wondering if converting back and forth may introduce other subtle problems

I recall reading somewhere that the conversion to/from the LAB mode is lossless. Using LAB as an intermediate mode may be a solution. I haven't tried or tested this, so take this with a big grain of salt. <g>

[hpycmpr]

Originally Posted by edgework

I do the same thing when I have to work in RGB. The rules are roughly similar, but keep in mind that Photoshop simulates the qualities of light in RGB, and ink on paper with CMYK. One significant difference is that in CMYK, applying a curve to any channel will affect both color and lightness in a very different way than with RGB. A rule of thumb is that you use the individual channel curves in RGB to adjust color, and you use the master curve to adjust contrast. Because Red, Green and Blue are all equal in intensity (equal values combine to produce dead neutral) the master curve can be used effectively to make lightness moves without altering the hue and saturation (although the lightness channel in lab mode is far more efficient). In CMYK, since cyan is such a crappy ink, it requires about 10% more than Magenta and Yellow to balance the other two and produce neutral. Using the master curve in CMYK is very dangerous, unless you are using it in luminosity mode.

Excellent points about the difference between rgb and cmyk. I also adjust individual curve channels first, and leave adjusting the composite at the end and only when necessary. Now that I'm beginning to correct portraits, it gives me motivation to try cmyk.

Originally Posted by edgework

I was trained to use CURVES, and CURVES only, (and only in CMYK) since most of the other adjustment tools are more or less subsets of what curves are capable of doing. However, I've found that using selective color in the way previously mentioned, is a great quick fix. If, for example, a face has yellow highlights and hot red shadows, pulling magenta and adding yellow to the reds, and pulling yellow and adding magenta to the yellows brings the tones into balance, albeit a flat kind of balance.

Curves is my main tool. After that, I will use other tools to do whatever I cannot do with Curves.
Curves pros:
- They can do a whole lot of things.
- A steep portion of a curve boost contrast.
- I kind of feel like "knowing" what is being done to the pixels.

Curves cons:
- Needs patience.
- Steepening a portion of a curve flattens another (losing contrast). Can't have my cake and eat it too!

Originally Posted by edgework

Another trick for skin (this is for CMYK corrections): I've never found a face that cannot be improved by making a contrast move to the magenta channel in luminosity mode. Flat skin tones are almost always the fault of the magenta plate, and even if the numbers are within a realistic range, the lack of contrast can make the skin appear too hot overall. Heightening the contrast in magenta, but keeping it in luminosity mode can open up detail in an instant, without messing up the color.

I'll definitely keep this in mind.

Originally Posted by edgework

As for converting back and forth, the areas of RGB that extend beyond CMYK are far more extensive in the Greens and blues. The truth is, any colors in RGB that are outside the CMYK gamut aren't going to look all that appropriate in skin tones anyway--greens, blues, hot reds and day-glo oranges. If realism is the desired goal, moving into CMYK before correcting the color will weed out a lot of unrealistic tones. But if intensity and bright colors are the visual goal, and the target is RGB output, or the WEB, keep it in RGB.

Good point. For printing on a desktop inkjet, this may even be a less issue.

Originally Posted by edgework

One problem with CMYK that is non-existent in RGB is ink density. Because there is a limit to the quantity of ink a press can handle without turning the result into mud, the CMY channels will usually flatten out to a single tone in the shadow range, shifting the detail information into the black plate. This can be a problem if you have shadows that carry a cast, either hot or cool. Moving an image into CMYK will usually turn the dark areas neutral, which will be a problem if the mood of the original art called for overall warmth or coolness. I often use selective color for these situations too, pulling cyan from black and adding magenta and yellow, or the other way around to cool them down. Pay attention to the darkest shadows as well as dark hair: it's easy for rich brunettes to become dingy blacks in the translation.

It took me a long time to figure out that the ink density is handled differently between a press printer (cmyk) and a desktop inkjet printer (rgb). On the inkjet, the ink density is NOT (solely?) controlled by rgb or cmyk values.

[Duv]

Originally Posted by Duv

I myself haven't a clue how to adjust skin tones using RGB. I've always converted to CMYK to do my skin correction and then back to RGB. There seems to be a lot more information on CMYK skin correction. Although I haven't noticed anything, I'm wondering if converting back and forth may introduce other subtle problems. The tones you acheived are spot on. I never considered using Selective Colors to do my correction. I'll have to give it a try. Can I ask why you adjust yellow and red as opposed to magenta/yellow/cyan and black?

Cheers
Dave

Well now, don't I feel a bit silly quoting myself. I reread Eismann and realized I could have the best of both worlds.

Duplicate your RGB image
Convert to CMYK and make your adjustments
Drag it back onto the RGB layer and change blend to Color to bring back original RBG luminosity values

I've tried it quite a bit and it works like a charm.

Cheers

Dave

[hpycmpr]

Originally Posted by v.bampton

Ok, nice and easy...

If you're working on a decently exposed image, then you shouldn't have a problem. Find the blackest point in the image (that's supposed to be black), and watch the info pallette as you adjust. Move the end point of your levels or curves individual channels that black point reads a neutral number, (ie. R5, G5, B5). The move to the highlights, find the lightest point that's supposed to be white (not a specular highlight) and make it, say R245, G245, B245. If you haven't got a true white and black point in the image then it's a bit more complicated, so practise on ones that do.

Once you've got a neutral white and black point, you're mostly corrected. See if there is anything else that's supposed to be a neutral midtone - maybe some ones suit, or church stonework. Beware that if the point you pick isn't actually supposed to be neutral, then your midtone colours will be thrown off. Again you're trying to get a neutral - where your R, G, and B number are all about the same.

Once your image looks (and is by the numbers) neutral, you can adjust for taste. I tend to add an additional adjustment layer on top for my personal adjustments, leaving the neutral adjustments as they are. That way if I don't like my adjustments I can always go back to neutral without having to start again.

I tend to add red and yellow to the highlights and midtones. Something like R 0, 1.05, 250, 0, 255. G 0, 1.0, 255, 0, 255. B 0, 0.93, 255, 0, 250. It's not exactly neutral, but adds a nice warm glow to the skin which for portraits is ideal.

Hope that helps a bit.

Correcting the neutrals is how I start with each image. An ideal image for this kind of correction is one with true neutrals at highlight, midtone and shadow. But as you pointed out, sometimes this is not as simple or even possible with some images. In order of difficulty, here are some examples:

Some images are shot in certain lighting conditions with a desireable color cast that should not be removed, such as landscape scenes at dusk or dawn. Whether these images have any true neutrals is irrelevant.

Some images do have true neutrals but they are difficult to identify, such as several statues with different shades of "gray" marble.

Some images do have true neutrals that are easy to identify, but are not at all three points of highlight, midtone and shadow.

An image shot in a studio under controlled lighting and the inclusion of a gray wedge will not have any of the above problems.

[McCombs707]

First off, Dan Margulis is a genius, and I think his work is incredible. But, wanting to just work in RGB myself (though I work in the commercial printing field, where it's all CMYK) for my own personal photography, which will be printed on Fuji Frontier or similar equipment via sRGB, here's my formula for skin tones, at least for caucasian kids (adults are very similar)....

This is really very simple, just takes a while to 'splain'.....

After adjusting highlights and shadows via Levels, take a reading using the eyedropper with a 5x5 pixel setting of a good area of skin tone, not rosy cheek, middle of forehead works well normally. Set the eyedropper tool to display a reading in the info pallette by clicking it in the toolbar while holding down Alt (PC only, I forget which key to use on a Mac right now).

Using Curves, watch the setpoint you set in the Info Pallette and adjust the individual RGB channels to match what's below.

*** Only adjust from the UPPER RIGHT point in the Curves dialog, NOT from the midpoint or any point between lower left and upper right! Just for now, trust me!

Green - keep this constant
Red = 112.3 % of Green
Blue = 97.7% of Green

After correcting skin tone color doing the above, convert to LAB mode, and adjust contrast in Curves selecting only the Lightness channel, while viewing all channels, and adjust contrast to taste. This allows contrast adjustment without altering color range.

After all the methods I've tried, this works best, especially if you want to stay in the RGB colorspace.

[jenjen]

Originally Posted by McCombs707

Using Curves, watch the setpoint you set in the Info Pallette and adjust the individual RGB channels to match what's below.


Green - keep this constant
Red = 112.3 % of Green
Blue = 97.7% of Green

After correcting skin tone color doing the above, convert to LAB mode, and adjust contrast in Curves selecting only the Lightness channel, while viewing all channels, and adjust contrast to taste. This allows contrast adjustment without altering color range.

After all the methods I've tried, this works best, especially if you want to stay in the RGB colorspace.

How exactly do you figure out the percentages? Green keep it the same right? Red 112.3% or green? How is that possible? I guess i'm calulating this all wrong. I've been trying to figure the whole skin color thing for weeks now. Thanks!

Jen

[Stroker]

convert to LAB mode, and adjust contrast in Curves selecting only the Lightness channel

Or...
Stay in RGB mode, Curves Adjustment Layer, and set to Luminosity.
Few other notes about this, but should be good enough for most.

[inskip]

Originally Posted by McCombs707

*** Only adjust from the UPPER RIGHT point in the Curves dialog, NOT from the midpoint or any point between lower left and upper right! Just for now, trust me!

Green - keep this constant
Red = 112.3 % of Green
Blue = 97.7% of Green

How do you determine where you ned to grab on the curve line? Can someone elaborate a little on the above formula for me? I'm not sure what
Red = 112.3% of Green means??? Keep Green constant??? Does that mean don't change it?

[Duv]

Originally Posted by inskip

How do you determine where you ned to grab on the curve line? Can someone elaborate a little on the above formula for me? I'm not sure what
Red = 112.3% of Green means??? Keep Green constant??? Does that mean don't change it?

If I understand his post correctly, on the sample Original I put a sampler marker on her cheek and got readings of 219, 148, 109. Based on his percentages I should take the green value(148) and multiply it by 1.123 which gives me 166 for red. For blue I take .977 of the green value (148) which gives me 144. Therefore, I need to adjust my curves to get 166, 148, 144. He is suggesting pulling the curve from the top right corner only..not from anywhere on the curve.
I posted a Corrected Image based on the percentages and I can assure you it's not a skin tone that I find plausible. Perhaps I don't understand and need further clarification but I honestly can't see how a fixed percentage of colors can give you what you want. It certainly wouldn't work for a dark skinned person.

Cheers
Dave

[inskip]

Well Dave, it may not be perfect but it's definately improved quite a bit.
He said NOT to sample from the cheek, but rather the forehead which is generally not as rosy. Don't know how much difference that would make...

I'm still unclear as to how you determine where to pull on the curve-in general.

[Cameraken]

Duv

I also had a go with McCombs707 method. You have read his steps the same way I did.
His last step is to convert to LAB and adjust the lightness.
My attachment is your picture plus the last step.

I don’t look too bad to me.

Ken

[fat0n3s]

I agree with Duv on this one.

The method that is posted above, is really only changing the white point of the image. If there was a color cast, only in the highlights, then this meathod might work. If the color cast was in the midtones and shadows, this really wouldn't do any good.

There is not a magic formula for correcting skin tone. If there were, photoshop's auto color would work everytime.

Here is a good tutorial on curves. Page 3 of this tutorial gets into color correction.

http://www.gurusnetwork.com/tutorial/curves/1/

The tutorial dosen't get into great depth, but it is a good start.

[Ken Fournelle]

Duv,

I use your method for duplicating the RGB image, converting to CMYK, correcting and then pasting back into the original RGB image.

But don't your convert back into RGB from CMYK before your paste the duplicated image back into the original RGB image?

We discussed this in the past, and I thought that was your methodology.

k

[Duv]

Hate to be an old fuddy duddy here but I think my correction and Ken's are worse than the original. It's impossible for this caucasian woman to have 50% cyan with 50% magenta. Take a reading off the whites of her eyes. They've got more blue and green than red. Does that make sense? The last time I checked I had a whole pile of blood vessels in my whites which would make red the predominate color.
I've posted one here using values of about 9/42/44/0 on the marker. Cheek, forhead, nose..but it should be higlight, no shadow, makeup or natural blush if possible. This isn't suppose to be definitive or anything but using known rough values of CMYK for caucasian skin gets me one heck of a lot closer than this percentage idea. IMHO.
Ken, When you have your corrected CMYK file..flatten it, select all, copy and paste back into the original RGB file. Then change the blend mode to color.
BTW, I've tried the percentage routine on a number of images will equally poor results so maybe the original poster is not explaining something.
Inskip, if you pull from the corners of Curves you'll get more extreme changes to highlights and shadows. If you want you're change to be more transitional, pull down along the curve. For example if you want to change from 196 to 176, run your cursor along the curves line. You will see the numbers change. Locate on the curve where 196 is, click on the curve and bend it up or down to get your 176 number.

Cheers
Dave