Flesh tone experts - help!

[Rexx]

Originally Posted by Duv

So the problem may be:
a) the woman really is lacking in color

I am not sure. I don't think so. But the lighting was absolutely horrendous

Originally Posted by Duv

b) the original picture is overexposed going in

It was, and as I've said before, all the "best" ones have applied some sort of technique that affects the Red channel in some way

Originally Posted by Duv

d) I'm a twit. This picture is just fine. She looks great.

You are. Me too. We're all pixel freaks around here

Originally Posted by Duv

One thing I do like is the way it easily adds Cyan to the mix which I was having trouble acquiring with curves and levels.

Cyan? Cyan! It drives me mad! I hate Cyan!

Originally Posted by Duv

With all the renditions submitted by the "experts" I can't wait to see a new picture of this lady from Vegard.

I just succumbed to peer pressure and called the lady in question, and I'll see her on Tuesday. She agreed to having her mug shot taken again, this time with flash, to get a reliable white balance. (I wonder what colour her hair is...) Hmmm, I might even get a good shot of her This time no 10D though, but my own A2.

Oh, and by the way, they've printed the photo and hung it on the fridge door

[Duv]

Originally Posted by Rexx

I just succumbed to peer pressure and called the lady in question, and I'll see her on Tuesday. She agreed to having her mug shot taken again, this time with flash, to get a reliable white balance. (I wonder what colour her hair is...) Hmmm, I might even get a good shot of her This time no 10D though, but my own A2.

If possible, could you also try to get a bounced flash shot? Direct on camera flash I find pretty nasty.

Dave

[Rexx]

Originally Posted by Duv

If possible, could you also try to get a bounced flash shot? Direct on camera flash I find pretty nasty.
Dave

Sorry Dave, I have no external flash for the Minolta I used to have a Metz for my Nikon FE2 (still have it), and bounced it all the time, but it will not work with the Minolta

[Rexx]

My young violinist friend (16) and I were practising last night, and we discussed the concepts of visual light being only one octave as opposed to the audible spectrum being at least ten. We also had some fun with the keyboard. As expected, he found the 7/10,000 difference to be "vast" After playing for a while, we had to stop, as it sounded "awful". One of his strings was at least 0.5% off (It is very cold here now, so his violin had been subject to some temperature change, and had possibly not yet fully reached room temperature the first time he tuned it.)

So much about hearing, but here's the interesting part:
He told me that in ear training classes at school, they have to stop after about half an hour, as everybody is exhausted and unable to hear the difference between anything at all. This doesn't happen if we're playing real music. The ear remains keen.

If we translate this to Photoshop, it may be interpreted as the difference between actual seeing and studying "artificial colour" as we do, sitting here squinting at the screen. In real life, with "live" colours, we have absolutely no problems at all, just as we had to stop playing because of an apparently minuscule error in the tuning.

I know that professional wine and liquor tasters chew biscuits between each sip to neutralize the taste buds. Should we do the same? Take frequent pauses and look at "live" colours for a while?

[saby]

Hi Rexx!

I think this picture had 2 problems:

gradation problem: it had no white point
colorcast problem : neutrals are disapeared

to solve the gradation problem U should set a white for the picture by the levels or the curves

colorcast problem all of the neutrals are colored so what i did that change the consist of neutral colors in the seletive color correction +7C -7Y +3K
skintone became ok for me some hue/saturation changeing i did too change the reds less saturated and the colorcast moved to yellow +7

[hpycmpr]

Thanks for the response.

Originally Posted by Duv

hpycmpr..good to hear from you again. I'm glad you plunked into this thread. On many levels it reaches to what retouching and correction is all about. Personally, I don't buy into this too yellow too red consideration by Ro. I'm sure I'm probably wrong. At least, in the original image I don't see too red. I certainly see (by way of the picker) greater red in the shadows..so what? Maybe I'm missing something (usually am) but most images I look at (see attachments) have higher red relative to the other colors in the shadows. So how is Ro coming to the conclusion that there is too much red in the shadows? Is it objective? Subjective? Is everyone seeing something that I'm not in the image? I'm feeling like an old, geriatric twit.

Unlike Rexx, we don't know what Lene really looks like, or under what lighting condition was the photo taken, or how the corrected image should look like. This probably explains why there is such a wide range of results and opinions. Correcting an image of "unknown source" can be either more or less challenging. For this image, we can probably agree that certain casts are unacceptable (to Lene at least), such as purple on the lips, or green in the hair. It would be interesting to present some samples to her and ask her which one she likes the best, and which one she thinks is most accurate (two different opinions).

Originally Posted by Duv

You're image certainly looks better to me after being Ro clipped but you could also come close by additional curve tweaking. There really is something to be said for working by the numbers and the KISS principle. Images with clear black, white (and even better grey) points are relatively easy to correct. Images like the one posted here is a combination of making sense of skin tone percentages, personal preference..and common sense. You've done a good job, maybe lacking a tad in Cyan ( personal only). Stay involved and don't be afraid to put yourself out there. We can all learn from you as well!

I rely on the numbers and KISS because I don't know many other more sophisticated methods, such as those in some tutorials. You hit the nail on the head about correcting images with true grays is relatively easy. Images which do not have true grays are definitely more challenging, and there are plenty of these. It took me a while to realise this, and Margulis spent a whole chapter on this topic. Books and tutorials usually focused too much on how to remove cast by fixing the grays, but not enough on how to locate or identify the grays to begin with. Correcting a false gray is the beginning of the end, regardless how good the tool is.

[Stroker]

Ven, you know me when I get the bug.

Rex, I gave a quick little description of how Doug's technique works.

Do photographers use the step-wedge thingie?
Grab a dowel, some paint, and step-wedge it.
Toss it into at least one photograph to get an idea of what corrections need to be done.
You could even do a dowel with RYGCBM.
Maybe I'll make a few and try to sell them on eBay...

[hpycmpr]

Originally Posted by byRo

I don't think this is a good idea. The reflection on the lips is not a "natural" white, the real colour at this point actuallly would be a red. As I see it, this point is a forced (blown-out) white which doesn't depend much on the actual lighting of the ambient. Thus you will be correcting what happened after taking the picture and not the lighting at the time of the picture.

Maybe the white of the eyes will do, but I haven't tried that.

Thanks for the feedback. I normally do not set white points on blown out whites. In this image, since there are no obvious black and white points (see my response to DU), I took the liberty to guestimate black in the pupil and white on the lip reflection. In this case, the blown out whites on the lip are small enough that they are not offensive. I intentionally left these untouched when applying your unclipping tutorial. Eye whites are typically not true whites, so I usually avoid setting white points there.

On a different subject, I have a question about your unclipping tutorial. After painting over the clippings, I have to restore the original contrast by applying the Brightness/Contrast tool a second time. Is there a way to get around this? Hate to spend a round trip on this tool.

[Duv]

Originally Posted by Rexx

She is an average Northern European woman, in other words, fair skin. Her husband says her eyes are a dark green, but the way the light comes here, I guess "any green" will do.

Well, We seem to be throwing a lot of darts and hitting the board but are we anywhere near the bullseye. We have a black point, the iris but we lack a true grey and white point. Perhaps working with the green in the eyes can get us closer. Given Vegard's comment above, the green in a lot of the renditions seems too light. Perhaps a Luminosity curve to darken the overall image or tying the yellow/magenta values together and increasing the percentages would result in a closer "bulleye" rendition.

Cheers
Dave

[Rexx]

Sorry folks, this has taken some time, but Lene and I and my camera had to coincide in space and time After that, I simply needed time to pp again. Well, now it's Easter holiday and here we go!

In the meantime I've attended a colour calibration course with GretagMacbeth, the "Picture Perfect Color Seminar". I learnt one very important thing there that applies to this "project". RGB or CMYK values for flesh tones are absolutely worthless unless you state which colour space they apply to!

Back to Lene. What do I have to work with now?
- The original shot in ISO 1600, with the Red channel blown all over the place.
- A fresh, slightly underexposed facial shot from my KM Diamge A2, in sRGB using flash. This makes me fairly confident that the colour balance is correct.
- A pivotal understanding of the image having different colour problems in the shadows and highlights. (Thanks byRo)
- An understanding that CMYK or RGB values are worthless out of context, read colour space.

On page 82 of the second edition (2003) of her book, Katrin Eismann shows a technique for restoring a blown channel. Now who am I to stand up against Eismann? Let us just say that I am not convinced that this technique will always yield satisfactory results. It corrects the lost detail of the blown highlights without changing the colour (would that be hue?) of the image. But with one channel already blown, doesn't that throw the colour balance off? So why preserve it? I will try my personal approach.

I started thinking fresh. I tried to imagine what the image "looked" like. Here is my visualization:

First. let us imagine a b&w photo. When we look at it, we see intensities that vary from min to max. Min represents total black and Max represents total white. Everyone knows this, but bear with me while I'm stating the obvious.

Now think again. This b&w photo can be compared to a (rugged) landscape that we're flying over at high altitude. White represents mountains and high areas; black represents valleys and low areas. If we lay a vertical slice through this landscape, we get a height profile, just like you see in Tour de France or downhill skiing or whatever. Or like in CT (Computer Tomography X-ray).

If we lay such a slice across Lene's forehead in the Red channel, we get a nicely profiled curve in the temple area, but when we reach the forehead proper, the curve turns completely flat! It will look like those funny table top mountains you have in the US deserts. The Red channel has reached the technically maximum height (255), and is unable to show the real profile. What should have been there, is completely lost. Or is it?

The Green channel is not blown. In addition, the Green channel is usually almost synonymous with the Luminosity channel. In fact, my camera uses only Green for contrast focusing. Green is our luminosity and contrast friend.

I can make a mask from the Red channel, white where the channel is blown (255), black for the rest. I can then apply this mask to the Green channel. What does this represent? If again we apply a vertical slice, but this time to the masked Green channel, we see the profile of the "Green mountain", covering the same "aerial view" where the "Red mountain" was flat. We can think of it as if the water rose to the level represented by the mask, and only the unmasked area remained as an island.

I have now a fairly good guesstimate of what the Red channel should have looked like in the area where it is blown. There is just one problem; I cannot make the "Red mountain" taller :-( But what about "lowering the terrain"? Going back to the masked Green channel, I can use the histogram to see how much taller the "Green mountain" is than the area outside of the mask, or in the island metaphor, how tall is the tallest mountain on the island? If I "lower the terrain" to the same degree in all the channels and apply the "Green island" profile to the blown part of the Red channel, I should have re-created an approximation of what a correctly exposed photo would have looked like. This will of course not allow for colour variation in the skin, since the ratio of Green to Red will be static and unnatural. However, I hope that Blue (which I haven't touched) will take care of that, since it is notoriously blotchy and ... lively?

The difference between this method and Eismann's is that I don't touch the part of the image that is not overexposed.

Now I hopefully have a correctly exposed image, and can start to have a look at the colours again. Gaussing Lene's forehead from the flash shot should give me a usable swatch sample of her skin.


Ok, here we go.

1) Import and convert from Adobe RGB (the colour space used in the 10D) to sRGB (my working colour space)
2) Create a mask from the Red channel. Set Threshold to 255. Store the mask as an alpha channel. I will need it later. Name it "Blown".
3) Create an Art layer from the Green channel, for instance using the Channel Mixer.
4) Apply the mask from above. We now have a strange, grey, ghostly shape of most of Lene's face. The Red channel was worse than I had thought.
5) Create a Black filled layer underneath this one.
6) Merge the grey ghost face down into the black layer. Now we have a ghost surrounded by black.
7) Show the histogram for this layer. It actually has values from about 150 to 225. That means the "Green mountain" is as much as 75 levels high.
8) 75 is 1/2 of 150 and 1/3 of 225. The "tallest Green island mountain" within the blown Red area is 50% higher than "sea level" at 150. It means that this image is overxposed by exactly one half f-stop. It needs to be pulled back 1/3.
9) Use the Channel mixer and set all three channels to 65%, about a half f-stop lower exposure. Now we have the levels where they should be, and there is room to re-expand the Red channel.

That was the easy part. I will not need any of the layers anymore, except the "Blown" alpha mask, so I can throw them away.

Now how do I partially replace one channel? And it should be seamless. Hmmm... I have this uncanny suspicion why Eismann prefers that ... other method

After some hours of trial and terror, I think I have something.

10) Working on specific channels is easier in Greyscale. Create a greyscale image, same size. Drag the Red and Green channels from the working document to the Greyscale document. Drag the Blown mask too.

11) The channels become alpha channels. Disappointment. I thought they might instantly turn into layers. Copy-paste into Layers, "Red" and "Green".

12) Create Levels clipping Adjustment layers for both the Red and Green layers. Adjusting the right slider for Green to 230 looks ok to me. Only a few reflexes reach 255. The Green layer receives the Blown mask.

13) For the Red layer I adjust the output level (the one at the bottom) to approx 175, then use the arrow keys to do fine adjustment. I end up at 176. This number depends on the number from the Green levels.

14) I Gauss the Green (Blown) mask with a radius of 3.0. Better.

15) Still some mismatch. I use a brush at 10% on the Green mask a few places. Left eyelid (her right), left cheek (her right), lip where it turns into shadow. I think I have a new Red channel!

Below the two versions. They uploaded in the wrong order, so the strong yellow one to the right is the original, and the darker one to the left has a reconstructed Red channel withut blown highlights.

The rest of the story in the next post, hang on...

[Rexx]

To create a good flesh tone swatch of Lene, I took my flash shot of her, Gaussed it at 50 pixels radius, and increased the exposure to a "sensible" level (flash reflection on forehead approaching 255). I then simply clicked around on her forehead, jotted down the readings in Excel and sorted them. NB! These are sRGB values.

R G B G/R B/R
164 113 93 69% 57%
165 116 95 70% 58%
171 123 100 72% 58%
179 123 103 69% 58%
180 124 104 69% 58%
191 133 110 70% 58%
192 135 114 70% 59%
206 145 123 70% 60%
209 152 130 73% 62%
217 161 139 74% 64%
225 169 150 75% 67%
235 181 162 77% 69%
238 180 162 76% 68%

Ahhh, this doesn't look good. The tabs are lost.

I then Gaussed my reconstructed Lene somewhat, and was actually able to place a control point that was both on her forehead and on the forehead of the flash shot. Then it was just a matter of correcting the Colour balance until the two images matched. I used the Channel mixer.

Ok, from the left, the hall where the first shot was taken. Both Lene and I were sitting on the bench along the wall. Second, the flash shot of Lene, Gaussed. Third, the reconstructed Lene, Gaussed to make adjustment easier. And finally Lene with the same colour values as the flash version.

This is where I lie down and cry

[Legacy~Art]

I am no expert lol but i do like to play photoshop challenge...

I used PSP9 to correct the colours, then i used curves to enhance the lips, fur, hair, then i used the patch brush...

[venivedi]

Originally Posted by Rexx

...
I learnt one very important thing there that applies to this "project". RGB or CMYK values for flesh tones are absolutely worthless unless you state which colour space they apply to!
...

I guess the seminar was so helpful. Thanks for sharing. You've reminded me of an important thing.
By the way, I see red cast in a previous posted image of mine.
So...just second try...here it is. What do you think?

[Hephaestos]

Well I tried my hack method again. Her left eye has a clear expanse on the cornea, so I used that as a gray point. Dropper showed red 252 (the whole red channel is pretty blown out), green 189, and blue 117.

So I adjusted the brightness in the color channels, red down 63, blue up 72. The high number of these adjustments is a giveaway that this method is not optimal for this kind of fix, but I tried my novice skills with levels and curves and came up with something even worse.

To me the lighting conditions are the real killer. I'm beginning to suspect that the necessary color information is just not there, in which case manual colorization in spots might be the way out.

[Rexx]

Originally Posted by Rexx

This is where I lie down and cry

When I set the flesh colour of an image identical to a known good shot of the same person, the second image looks awful

I should return to Beethoven instead. That would at least make my piano teacher happy...