Flesh tone experts - help!

[realaqu]

I am a newbie in here and learned a lot from you experts. my approach.
I usually check the rgb chanels. the red channel is too bright and blue is too dark. so I apply 30% green to both red and blue in normal mode to fake some details, and add 30% to green channel in multiply mode. then create a color balance adjustment layer to balance skintones. then another curve adjustment layer to adjust contrast. here is what I got.

[Hephaestos]

I like those results realaqu, I'm writing that down.

[realaqu]

Thank you Hephaestos
I forgot to mention two things I did. I use dust and scratches filter on blue chanel to reduce some noise. and I found there is some yellow on her nose. so I use layer style to blend it away.

[barry_uk]

I have this thing about colouring faces called fascination LOL i could'nt help myself but have a dabble myself all opinions I m open to criticism whats life with out it LOL been married 20 yrs I am use to it by now HAHA. "Dont we just love women boys "

[BayCanuck]

I thought I'd give this one a try. I made about six attempts, based on lots of ideas found in the thread. Here's my "final" version.

1. Use the "match colour" trick on a new layer & blend in colour (thanks for that one!)

2. Set four colour samplers in the skin.

3. Converted to CMYK

4. Copied the yellow channel, then used levels to push in the white and black points.

4a. Loaded the new channel as a mask

4b. Added a curves adjustment layer and did a "by the numbers" correction on the four samplers (Y ~= M + 2)

5. Copied the magenta channel, then used levels to push in the white and black points by eye

5a. Loaded the new channel as a mask

5b. Added a curves adjustment layer to increase the contrast in the magenta channel.

6. Added a levels adjustment layer for the cyan channel (12, 0.88, 288)

7. Played with the layer opacities of the adjustment layers. The final values were: Yellow 69% opacity, Magenta 85% opacity, Cyan 100% opacity

Cheers,
-Darren

[realaqu]

Originally Posted by BayCanuck

I thought I'd give this one a try. I made about six attempts, based on lots of ideas found in the thread. Here's my "final" version.

1. Use the "match colour" trick on a new layer & blend in colour (thanks for that one!)
rs. The final values were: Yellow 69% opacity, Magenta 85% opacity, Cyan 100% opacity

Cheers,
-Darren

Hi,
I am not sure if that is my monitor's problem or sth else, your picture looks a little bit greenish on my machine.
My own thoughts when I started to fix the picture. is this is a typical indoor bad shot. The reason, The CCD or CMOS sensor in DSLR just simulate the traditional RGB layered film camera. the blue layer is at the bottom of the film base. when green layer get a correct exposure. the blue layer still doesn't get enough lights to make correct exposure. that is the reason why the picture lost so many dark details in blue chanel.
So I myself think there is no way to get a perfect exposure in this case to fix that picture except we fake some details in both damaged red chanel and blue chanel. after this step, I can use the regualr methond to play with levels. curves or whatever neccessary steps to correct the color things.

just a newbie's opinion and please do excuse my bad English. hope you can understand what I am saying.

[Duv]

It's very hard to disagree with you Realaqu! I personally don't agree with the Idea that the Reds are totally blown out or that the blues have no dark detail. If you take a look a the Luminosity channel in LAB mode, things don't look that bad. Obviously, there are a lot of different approaches. I got what I thought was pretty good results copying an improved LAB luminosity adjustment curve and pasting into the Red Channel. Also as suggested before, running a modest noise reduction on the blue layer plus even copying some LAB luminosity into it can give you a strong footing for your color adjustments.

Also, don't forget to think about Color Profiles for corrections. I have over 50 with different temperatures and gamma points that I often use for corrections. I'm really beginning to think that for a lot of corrections, especially badly under exposed images, Gamma point change is the way to go as no banding occurrs as with curves and levels adjustments.

My own opinion is that the Red channel ain't that bad..blown out a bit in the quarter tones but easily fixed. The real problem has always been the blue channel with nothing in the quarter or half tones. IMHO.

Not sure, but to my Canuck friend, on my monitor, I'm showing almost equal amounts of Cyan to Magenta and Yellow. May want to check that.

Cheers
Dave

[realaqu]

Hi, Dave,

The original picture does look good in LAB mode and your way works great, the reason why I use my methond to correc the exposure is because the color space of digital cameras, scanners are all device dependent. A picture with perfect exposure, the RGB curve should start or end at the same point. I can clearly see the blue chanel is too dark from attached histogram screen copy of the original picture. which shows we lost a lot of details in bright area of blue chanel. although you changed the display mode, but the lost details still can't be saved back. only the green chanel get correct exposure. that is the reason I use the green chanel to fake some details in blue to make the balance. I thought that might be much easier to play with color things. I double checked the red chanel, it seems ok actually. only lost a little details in dark area. seems I don't need to put some green on that chanel.

realaqu

[byRo]

Originally Posted by Duv

If you take a look a the Luminosity channel in LAB mode, things don't look that bad. .... I got what I thought was pretty good results copying an improved LAB luminosity adjustment curve and pasting into the Red Channel.

Dave,
Luminosity is made by mixing the three channels together. The weights are Green 59%, Red 30%, Blue 11%.

There are some differences of opinion about the Red and Blue channels (*) in this photo, but it is clear that the Green channel is OK. As luminosity is mostly made of Green then it follows that the Luminosity will also be pretty good. (in realaqu's histograms you can see that the luminosity histogram is pretty much a copy of the Green histogram)

In fact, you'd probably have done better to use the Green channel to fix the Red (**) and not the Luminosity.

(*) The Red channel is totally blown-out in half the photo and that is the big problem.
(**) Using the Red for the low end and Green for the high end - later I'll post the method.

Rô

[byRo]

I would like to reconsider something I wrote way back in this thread. The highlight / shadow colour differences that I had detected are, in fact, just symptom. The real problem here is a completely blown-out Red channel.

Using Rexx'x analogy, the way to fix it is to detect where the Red channel "moutain" went flat and slice in the corresponding part of the Green "moutain".

Here's how to do it:

1) Separate the Red channel into a layer on its own: (Channel Mixer layer, monochrome, Red 100%; Merge to this layer <ctrl><shift><alt><E>; name this layer "Red";

2) Mask out the blown-out parts: Threshold layer; put the triangle where the histogram gets distorted [245] (see photo 1); Make a luminosity selection <ctrl><alt><~>; invert the selection <ctrl><shift><I>; Select the "Red" layer and Add Layer Mask; discard the Threshold layer;

3) Mark the transitions: Using the eye-dropper <I> put in 4 markers <shift><click> at the edge of the mask - choose regions of smooth transition, not sharp edges / shadows (see photo 2);

4) Separate the Green channel into a layer on its own: (Channel Mixer layer, monochrome, Green 100%; Merge to this layer <ctrl><shift><alt><E>; name this layer "Green";

5) Make full-scale Green: Apply levels <ctrl><L> to the "Green" layer so that the lightest parts are at 255 [input levels 0,1,225];

6) Note the transition points: Using the info palette <F8> note the average value of the 4 marker points [170];

7) Adjust the "Red" to the "Green": Using the "Red" layer, deselect the mask <shift><click on the mask>, Levels <ctrl><L> adjust [output levels 0,176] until the info layer shows the maker points with the same number you noted earlier (170);

8) Combine "Red" and "Green" to make the new "Red" layer: Turn the mask back on <click on the mask> and the "Green" shows through. With the mask still selected run a Gaussian Blur to smooth out the transitions [radius 5]; Merge the "Green" and "Red" layers (turn off Background, <ctrl><E>, turn Background back on) (see photo 3);

9) Use the new "Red" channel: Open up the blending options of the "Red" layer and deselect the G and B options (see photo 4); we now have the original Green and blue channels from the Background and our brand new Red channel;

10) With the BIG problem solved a simple curve adjustment will fix the colours (see photo 5).

(I think I just wrote a tutorial!)

Rô

[realaqu]

Hi, Ro,

That is a great trick to repair the red chanel. but I still have a question here, I cropped part of your picture and read the histograms in photoshop. seems the curve still not that good. especially the blue chanel. you can take a look on my work. the RGB curves are smooth, almost start or end at the same positon. which means the exposure of that picture get corrected. maybe you need some more work on blue chanel as well.

[byRo]

Originally Posted by realaqu

..the RGB curves are smooth, almost start or end at the same positon. which means the exposure of that picture get corrected...

I think you are tending towards an auto-level adjustment <ctrl><L>, this will indeed improve the exposure but at the cost of colour. If the R / G / B histograms all start and end at the same position then things are getting pretty grey. It is perfectly normal, with skin tones, for the Red to be concentrated at high levels, green in the middle and blue only at low levels.


As it is pretty common to have to deal with blown Red channels, and as the method I posted above looks pretty complicated - I made a action toolkit to make this easy (well easier):

1) Channels to layers - makes 3 layers, Red / Green / Blue;
2) Combine channels - recombines the channel information;
3) Greyscale channels - makes the 3 layers greyscale;
4) Red fix - does all the work.

Run Channels to layers, then Red fix.
(and then curves to get the colours right)

Rô

[realaqu]

Hi, Ro
I don't meant the RGB curves have to be at the same starting or ending point. but for a good exposure, they have to cover the film dynamic range as wide as they could. that is the reason I said they almost start or end at the same position. I got this from

http://luminous-landscape.com/tutori...stograms.shtml

from your histograms, the area under the green curve are the total amout of geen lights of this shot. which means the picture is a kind of greenish. I worked on a similar case before, the photo is from another forum for discussion usage.
you might wanna have a try

http://www2.photoshopcn.com/attach/m...GgI7Z9ymXk.jpg

I am not sure if there is a copyright issue because somebody asked for help on that forum. so I didn't attach the original picture. but I uploaded my own work on that.

[Duv]

Originally Posted by byRo

Dave,
Luminosity is made by mixing the three channels together. The weights are Green 59%, Red 30%, Blue 11%.

I'm sure I'm missing something but this is precisely why I didn't use the green channel. By Blending in the green with the red and/or blue channel, you have weakened the green channel. How can it be otherwise. With your corrected version, LAB brightness (no mixing here I think) is now stronger than green channel so it might make sense to blend LAB back in to get rid of the apparent red cast to your image and put some strength back in the green channel. Also, when I blend the LAB "brightness not Luminosity" channel into the red channel, I think i get a better result. Also, here I paraphrase someone else
"Lets call RGB, LAB a color models as color spaces are used in the context of color management.

Sorry to spoil the party: RGB values only do not define any color. WOW?? RGB inside a color space does define an absolute color. Same RGB can mean very different colors in different color spaces. Any LAB color always defines an absolute color.

To me RGB is the least intuitive color model. The best for my taste is HSB as used in the color picker. But the mathematics seem to be against HSB and LAB models seem to be better.

If you think in terms of colors HSB rules:

1. Brightness (mostly responsible for detail in our vision system)
2. Hue (color base tone)
3. Saturation

In many tools and in layers it is good to think in brightness (or Luminosity not really the same but for me it is) and color (Hue, Sat).
Uwe

____________________________
Uwe Steinmueller
Editor Digital Outback Photo

Comments please Roland. I'm not as eloquent as you but still need to understand.

Cheers
Dave

[byRo]

Originally Posted by Duv

I'm sure I'm missing something but this is precisely why I didn't use the green channel. By Blending in the green with the red and/or blue channel, you have weakened the green channel.

Dave, couldn't follow you here. Seems that this is what I was saying. The Luminosity will have some of the distortion from the Red channel and some of the noise from the Blue channel mixed in - and that's why I'd use the Green channel by itself.

Originally Posted by Duv

To me RGB is the least intuitive color model. The best for my taste is HSB as used in the color picker. But the mathematics seem to be against HSB and LAB models seem to be better.

If you think in terms of colors HSB rules:

1. Brightness (mostly responsible for detail in our vision system)
2. Hue (color base tone)
3. Saturation

In many tools and in layers it is good to think in brightness (or Luminosity not really the same but for me it is) and color (Hue, Sat).
Uwe

____________________________
Uwe Steinmueller
Editor Digital Outback Photo

Fully ageed! I was lucky enough to cut my Retouching baby teeth on PSP and not PS. In PSP the HSB is (or was, anyway) a front-line option and not something relegated to the off-line "Goodies" folder of the CD. I will always prefer treating Luminosity and Colour separately.

Originally Posted by Duv

Comments please Roland. I'm not as eloquent as you but still need to understand.

Dave, it's content and not the packaging that counts! ah... and don't follow me - I'm not sure where I'm going

Any how, the point that I was trying to make os that once you've fixed the blown-out Red channel the rest of the correction is pretty straightforward.
I have a little collection of skin Gradient Maps that I use for corrections / colourizing - here are three examples of correction: Blushing red, normal(?), Latin yellow.

Rô

(PS the numbers on the attachments are Hue and Saturation at 60% Luminosity)