Which skin tone is best? Four different profiles

[ErikKaffehr]

Hi,

I looked at some other images. Tried to process a P45+ image in Lightroom with my QPCard generated profile and C1. I came up with the enclosed images, Capture 1 left LR5.3 Right.

LR5 is essentially default, WB on wall left of head.

C1, linear response curve, WB on wall left of head, some adjustments

Best regards
Erik

[torger]

A bit late to the discussion, but anyway; I don't find the "which skin tone is best?" question particularly hard to understand, it's clear that what we strive for here is a natural day-to-day look, what you'd like to have out of camera when you've done no particular post-processing except for selecting a portrait profile. Not what you'd like to have after hours of post-processing for a magazine cover. Medium format has quite some reputation for being good at achieving pleasing skin tones "out of the box".

A pleasing natural skin tone in this case is probably a bit warmer than "accurate", and what would seem most natural to me on my uncalibrated screen I'm sitting at now is something inbetween the top left (LR5.3 color checker) and bottom right (capture one). As I've noted in some other threads I don't think you should use linear curve in capture one for best performance of their ICC profiles.

[Bart_van_der_Wolf]

A bit late to the discussion, but anyway; I don't find the "which skin tone is best?" question particularly hard to understand, it's clear that what we strive for here is a natural day-to-day look, what you'd like to have out of camera when you've done no particular post-processing except for selecting a portrait profile. Not what you'd like to have after hours of post-processing for a magazine cover. Medium format has quite some reputation for being good at achieving pleasing skin tones "out of the box".

A pleasing natural skin tone in this case is probably a bit warmer than "accurate", and what would seem most natural to me on my uncalibrated screen I'm sitting at now is something inbetween the top left (LR5.3 color checker) and bottom right (capture one).

Hi,

I agree with those observations.

As I've noted in some other threads I don't think you should use linear curve in capture one for best performance of their ICC profiles.

I'm not convinced as to the magnitude of the effect on color, yet. The Linear versus Film curve response is IMHO (or it should be) a tone curve correction that is more than a post processing curves adjustment. It should be a Luminance based tone curve adjustment that leaves color (shift and saturation) mostly unaffected. A great example of how it should work is the Intellicolor technology from Topaz Labs.

I'll do some testing myself with Capture One, because Linear response reduces clipping of correctly exposed files and maintains highlight contrast instead of compressing it, and that's why I use it almost exclusively. I'd appreciate it if you have some examples you can share that show the magnitude of the effects on color (but I'll do my own testing as well, so it's not that I'm lazy).

Cheers,
Bart

[Steve Weldon]

It is subjectivic but you asked what I like and I can  answer that..                                                                                                                                                                                                                                                                                                                              

The bottom left (opcard) seems the most pleasing to me.  The two on the right.. the reds are over saturated and the skin tons greyish...  

[torger]

I'm not convinced as to the magnitude of the effect on color, yet. The Linear versus Film curve response is IMHO (or it should be) a tone curve correction that is more than a post processing curves adjustment. It should be a Luminance based tone curve adjustment that leaves color (shift and saturation) mostly unaffected. A great example of how it should work is the Intellicolor technology from Topaz Labs.

I'll do some testing myself with Capture One, because Linear response reduces clipping of correctly exposed files and maintains highlight contrast instead of compressing it, and that's why I use it almost exclusively. I'd appreciate it if you have some examples you can share that show the magnitude of the effects on color (but I'll do my own testing as well, so it's not that I'm lazy).

Actually I think it's a plain RGB curve in Capture One. But I'd be glad to be corrected on that. The reason I think it's a RGB curve is that the TIFFTAG_TRANSFERFUNCTION when export TIF with embedded camera ICC profile indicates that.

DNG/Lightroom has a special type of curve as defined in the DNG standard, which is supposed to have "film-like" increase of saturation when you increase contrast with it, but it's also prone to color shift. A pure luminance curve I think most users would consider looking too gray and dull, although color stability is good. Most expect color to intensify when contrast is increased.

RawTherapee is a great raw converter experiment with different curves. It has Luminance, RGB, Adobe's film like, and their own "weighted standard" which is my personal favourite, which is quite close to Luminance but not quite.

I don't necessarily think it's wrong with a simple RGB curve with color shift and all, as long as you design your ICC to work on that curve. If you want color stability with total freedom to select curves my guess is that none of the current commercial raw converters are particularly good at that. It has not been a design target for them, the workflow for "accurate" color is to make a custom profile tailored for a specific use case, ie curve and illuminants chosen and exposing correctly.

The reason I got into this issue through working with Lumariver HDR (we now in version 1.1.1 support Adobe's DCPs natively, and I'd like to support Capture One ICCs too to make it more flexible in MFDB workflows) and also contributing to RawTherapee (the upcoming 4.1 release will hopefully have support for Capture One and Leaf ICCs) both applications which find tone curves applied through/with profiles a bit problematic as they're designed to start with linear input.

For the luminance curve to work as you intend the ICC profile must be applied before the curve is applied, but that is certainly not the case in Capture One. As said with Adobe's DCP it is, but then their curve is only marginally more neutral than a plain RGB curve.

I have not made any screenshots, there's some work to do it so I'll see if I get around to make some.

[ErikKaffehr]

Hi,

As far as I know, LR uses something called Melissa RGB internally, it is essentially ProPhoto RGB with gamma = 1. I believed until this day that a camera profile would map RGB channels to XYZ or Lab in Melissa RGB. Anyway that is what I feel a camera profile should do. Rendering, that is a different thing.

Best regards
Erik

Actually I think it's a plain RGB curve in Capture One. But I'd be glad to be corrected on that. The reason I think it's a RGB curve is that the TIFFTAG_TRANSFERFUNCTION when export TIF with embedded camera ICC profile indicates that.

DNG/Lightroom has a special type of curve as defined in the DNG standard, which is supposed to have "film-like" increase of saturation when you increase contrast with it, but it's also prone to color shift. A pure luminance curve I think most users would consider looking too gray and dull, although color stability is good. Most expect color to intensify when contrast is increased.

RawTherapee is a great raw converter experiment with different curves. It has Luminance, RGB, Adobe's film like, and their own "weighted standard" which is my personal favourite, which is quite close to Luminance but not quite.

I don't necessarily think it's wrong with a simple RGB curve with color shift and all, as long as you design your ICC to work on that curve. If you want color stability with total freedom to select curves my guess is that none of the current commercial raw converters are particularly good at that. It has not been a design target for them, the workflow for "accurate" color is to make a custom profile tailored for a specific use case, ie curve and illuminants chosen and exposing correctly.

The reason I got into this issue through working with Lumariver HDR (we now in version 1.1.1 support Adobe's DCPs natively, and I'd like to support Capture One ICCs too to make it more flexible in MFDB workflows) and also contributing to RawTherapee (the upcoming 4.1 release will hopefully have support for Capture One and Leaf ICCs) both applications which find tone curves applied through/with profiles a bit problematic as they're designed to start with linear input.

For the luminance curve to work as you intend the ICC profile must be applied before the curve is applied, but that is certainly not the case in Capture One. As said with Adobe's DCP it is, but then their curve is only marginally more neutral than a plain RGB curve.

I have not made any screenshots, there's some work to do it so I'll see if I get around to make some.

[Vladimirovich]

As far as I know, LR uses something called Melissa RGB internally

only to disply histogramm - otherwise it is the same internal workflow as ACR

[Vladimirovich]

Melissa RGB internally, it is essentially ProPhoto RGB with gamma = 1.

it is ProPhoto coordinates + sRGB curve

[Fine_Art]

A great example of how it should work is the Intellicolor technology from Topaz Labs.

Separating colour from luminance seems very logical. There was a scanner software that had a colour wheel for removing film casts. Being able to pick a point on the image, then change hue and saturation with a colour globe, while leaving or separately adjusting luminance would remove the mass of obfuscating sliders. It would probably clean up the code to far fewer bugs as well.

[Vladimirovich]

Separating colour from luminance seems very logical.

so does hue from saturation... ACR now has an option for Lab color samples (if you select output to Lab), may be they will treat us to some HSL/HSB color sample and an additional option to set it by selecting an arbitrary rectangle with color sample tool and displaying some average values within it (like it was finally done w/ WB tool)

[torger]

Problem is that color science is not an exact science, as it's about human perception. Lab, or HSV or whatever only approximates separation of hue, saturation and luminance. It's indeed a quite good approximation but it's not perfect. Even if it was, the perception is slightly altered when the curve is changed. Increased contrast in the luminance channel alone most will perceive as increased contrast *and* a reduction in saturation. Thus good pleasing color rendition is a little bit of black magic.

[Bart_van_der_Wolf]

Increased contrast in the luminance channel alone most will perceive as increased contrast *and* a reduction in saturation. Thus good pleasing color rendition is a little bit of black magic.

Hi,

Well, not entirely black magic. The change of contrast can be followed by a change in saturation to taste, and you're there, as the Intellicolor technology example shows. All it takes is a good separation between chrominance and luminance.

Intellicolor technology is also used in the incredibly powerful Topaz Labs Clarity plugin, also in a separate HSL control with edge-aware masking capability which makes it very useful for skin adjustments. Skin color, regardless of race, tends to have a very narrow common Hue angle.

Cheers,
Bart

[torger]

Well, not entirely black magic. The change of contrast can be followed by a change in saturation to taste, and you're there, as the Intellicolor technology example shows. All it takes is a good separation between chrominance and luminance.

Intellicolor technology is also used in the incredibly powerful Topaz Labs Clarity plugin, also in a separate HSL control with edge-aware masking capability which makes it very useful for skin adjustments. Skin color, regardless of race, tends to have a very narrow common Hue angle.

I think I actually shall take a look on that Intellicolor :-). I do think the approach is interesting and is what I intend to evaluate a bit further in my own work. I don't think either Capture One or Adobe has chosen the best paths concerning color rendition. AFAIK Capture One has quite some legacy RGB processing in their pipeline, and Adobe has by design chosen a special type of RGB curve they think is "film-like and pleasing", ie the color shift should be there because it's pleasing. I can't say that their design choices are wrong, just different. I do think it's unfortunate that Adobe has embedded their choice of curve, ie their view of color, in the DNG standard which I think weakens it as a standard. DCP profiles that contains a curve is by the standard required to render it the Adobe way. Fortunately you don't need to embed a curve at all in the DCP.

One reason I'm using RawTherapee is that you there indeed can start with something neutral, separate contrast from hue and saturation and just as you say increase contrast and then apply saturation as you like. However that's a "new" digital way of understanding and working with images, what Capture One and Adobe presents is a more of a "film-like" behavior, which makes sense when people still remember how film cameras worked. I hope though that a raw converter that thinks digital photography all the way gets larger penetration in the future, after-all many photographers today haven't even used a film camera so I don't think we should be stuck with "film simulators" forever.

[Bart_van_der_Wolf]

I think I actually shall take a look on that Intellicolor :-).

Hi,

By all means, do yourself a favor. The Clarity plugin alone is a must have, the technology behind it opens a lot of possibilities for various unexpected uses as well. For portraits/skin, a little negative "Micro Contrast" will reduce the visibility of pores and small lines, and the effect can be restricted to skin tones with a Color Range mask (apply a few times to catch all skin tones), and that leaves eyebrows and hair alone. A second run with the filter can boost micro contrast for everything else, like eyes.

The HSL sliders for Orange probably have most effect on skin-color and e.g. a little Luminance reduction will give a nice sun tan on Caucasian skin without turning it orange. Adding or reducing a bit of red saturation can simulate/reduce sunburn or other reddish discoloration. Asian skin can be easily lightened (which seems popular) without bleaching crème. Dark/black skin can also be adjusted without causing saturation or color shifts.

I do think the approach is interesting and is what I intend to evaluate a bit further in my own work. I don't think either Capture One or Adobe has chosen the best paths concerning color rendition. AFAIK Capture One has quite some legacy RGB processing in their pipeline, and Adobe has by design chosen a special type of RGB curve they think is "film-like and pleasing", ie the color shift should be there because it's pleasing. I can't say that their design choices are wrong, just different. I do think it's unfortunate that Adobe has embedded their choice of curve, ie their view of color, in the DNG standard which I think weakens it as a standard. DCP profiles that contains a curve is by the standard required to render it the Adobe way. Fortunately you don't need to embed a curve at all in the DCP.

I agree, their approach towards RGB color is somewhat traditional, although Capture one does offer some useful skin-color targeted controls.

One reason I'm using RawTherapee is that you there indeed can start with something neutral, separate contrast from hue and saturation and just as you say increase contrast and then apply saturation as you like. However that's a "new" digital way of understanding and working with images, what Capture One and Adobe presents is a more of a "film-like" behavior, which makes sense when people still remember how film cameras worked. I hope though that a raw converter that thinks digital photography all the way gets larger penetration in the future, after-all many photographers today haven't even used a film camera so I don't think we should be stuck with "film simulators" forever.

Indeed, RawTherapee can handle almost every situation with a bit of practice, but TL Clarity is unique (e.g. color neutral contrast adjustments, great color adjustments, and powerful intelligent masking), and a huge timesaver.

Cheers,
Bart

[G*]

I find this thread really important - and not too subjective for a broad discussion at all. Sure, you might push the skin tone to marsian-green, if it helps with your "concept", but what happens with the rest of the image? (Imagine somebody wants to sell the incredibly red t-shirt and the beautiful green model.) So consistency and something like "realistic" colour in the first place is pretty important from my point of view.
For me it’s pretty hard to chose from the four examples a) because I feel they’re too small, and b) because I know neither the guy nor his shirt nor anything else in the frame. But I would say that none of the renderings is too unrealistic in respect to the model’s skin.
The shirt is a different case. Maybe one can’t say much due to its color outside of the sRGB-file’s gamut, but just for the fun of it I opened the file with RPP, did auto-exposure and read the WB from a small area of the white window. What I got was a much more yellow shirt which I can actually imagine being sold at a retailer, whereas the shirt on the OP’s renderings does not really look à la mode with its strange blue tint. Did he really buy such a shirt? :-)
Maybe the OP can help to unravel this mystery …

[Vladimirovich]

if that's not a fashion shooting, but rather a casual portraiture then who cares about the precise exactness of the shirt color... skin tone that OP will be pleased with is way more important, even at the expense of the shirt, is it not ?

[ErikKaffehr]

Hi,

The color of the shirt is slightly outside Adpbe RGB, so you need Prophoto RGB to show it exactly. Lab coordinates under D50 illuminant are:  42,69,39

Regarding the guy I cannot give data at the time the picture was taken, but right now it is about: Lab 56, 18,14.

Both measured by ColorMunki Photo and calculated by PatchTool from spectral data.

Shooting conditions were hazy Swedish midday sun early july, drinking coffe.

Best regards
Erik

I find this thread really important - and not too subjective for a broad discussion at all. Sure, you might push the skin tone to marsian-green, if it helps with your "concept", but what happens with the rest of the image? (Imagine somebody wants to sell the incredibly red t-shirt and the beautiful green model.) So consistency and something like "realistic" colour in the first place is pretty important from my point of view.
For me it’s pretty hard to chose from the four examples a) because I feel they’re too small, and b) because I know neither the guy nor his shirt nor anything else in the frame. But I would say that none of the renderings is too unrealistic in respect to the model’s skin.
The shirt is a different case. Maybe one can’t say much due to its color outside of the sRGB-file’s gamut, but just for the fun of it I opened the file with RPP, did auto-exposure and read the WB from a small area of the white window. What I got was a much more yellow shirt which I can actually imagine being sold at a retailer, whereas the shirt on the OP’s renderings does not really look à la mode with its strange blue tint. Did he really buy such a shirt? :-)
Maybe the OP can help to unravel this mystery …

[tho_mas]

To my eyes none of the versions showed in the first post are satisfying.
The first 3 versions show clipped colors (crushed colors and missing differentiation in the shirt). They also look too contrasty and a bit artificial.
The C1 processing doesn't look that defective, however, it looks a bit lifeless.

Here's my quick and dirty version (processed on my laptop… not with my primary display).
To me, this looks healthy. Took me 3 or 4 minutes. From here on you can further tune gradation/contrast (saturation or whatever you want…).

Basics:

Input profile: P45+ Outdoor Daylight
Curve: Film Extra Shadow

WB: 5000K / -1,8
Exposure: +1
Brightness: +6
Saturation: +10

2 Edits in the Color Editor: one for the shirt, one for skin tone (see attachments). I think you can't completely separate the shirt from red tones also contained in the skin tones so you have to find a compromise.

You can also try the "Portait Neutral"" Input Profile as a starting point ...

[ErikKaffehr]

Hi,

The images I posted were not processed at all, just applied colour profiles and adjusted exposure to match brightness. I perhaps did not make that clear. On some the shirt is clearly oversaturated. The red is a bit outside Adobe RGB (measured on the shirt).

Thanks for suggestions on processing. Good advice as usual!

Best regards
Erik

To my eyes none of the versions showed in the first post are satisfying.
The first 3 versions show clipped colors (crushed colors and missing differentiation in the shirt). They also look too contrasty and a bit artificial.
The C1 processing doesn't look that defective, however, it looks a bit lifeless.

Here's my quick and dirty version (processed on my laptop… not with my primary display).
To me, this looks healthy. Took me 3 or 4 minutes. From here on you can further tune gradation/contrast (saturation or whatever you want…).

Basics:

Input profile: P45+ Outdoor Daylight
Curve: Film Extra Shadow

WB: 5000K / -1,8
Exposure: +1
Brightness: +6
Saturation: +10

2 Edits in the Color Editor: one for the shirt, one for skin tone (see attachments). I think you can't separate the shirt from red tones also contained in the skin tones so you have to find a compromise.

You can also try the "Portait Neutral"" Input Profile as a starting point ...

[tho_mas]

The images I posted were not processed at all, just applied colour profiles and adjusted exposure to match brightness.

I see. In this case it's pretty clear that C1 provides the best starting point as it doesn't overemphasize certain colours. As I said the C1-image looks a bit lifeless by default... but it's balanced.