#1 portrait problem -- that magenta cast

rutt

Gmonkeh posted a shot on the Whipping Post. Here it is side by side with my edit.



I think it should look more like my edit than the original. In fact, unless you are looking for some sort of very special effect, I know it should look more like that. If you print the former, you will surely be unpleasantly surprised.

How do I know? I know because I measured the color values of her skin, especially on her forearms and chest (where there is unlikely to be makeup) and found that they are much more magenta than yellow. Long experience has taught us that nearly 100% of viewers don't like this compared with the opposite except in very special circumstances (sunburn, rosy cheeks, outside in a snowstorm, candlelight.) Certainly not for beauty. This is so well understood a phenomenon and so often reinforced by new data as to be close to science instead of an artistic choice.

I corrected the orginal with a simple change to the blue curve:



(There are many other ways to do this, but this one is should be a basic part of everyone's toolbox.) What's going on here? Yellow and blue are opponent colors. Yellow pigment doesn't reflect blue and blue light doesn't reflect from yellow surfaces. This curve reduces the blue through the low midtones where the flesh resides. Reducing the blue, comes to the same thing as increasing the yellow, allowing better balance with the magenta.

So much for theory. You don't actually need to know it. Bring up RGB curves, select the blue channel. Hold down the mouse button and mouse over the relevant parts of the image. You will see where they lie on the curve. Read the numbers from the info palette. In LAB, A should be no greater than B and both positive. In CMYK, M should be no greater than Y. In RGB, green should be a little higer than blue (same thing as Y being greater than M.)

There is a great tutorial on this by Baldy.

Still with me? I have to say something here that I have said many times before. DON'T TRUST YOUR EYES. DON'T TRUST YOUR CALIBRATED MONITOR. JUST SAY NO. The reason? Your visual system really really wants to correct for ambient light and show you correct skin tones for healthy people. Since this is obviously a healthy young woman, you will tend to see a healthy skin tone here, at least on your monitor which is relatively more immersive than a print. As you are the photographer and have looked at this image quite a bit, this will be even more true. Moral: Always measure the color values of skin tones. It's more important to get them right than any other single aspect of color.

pathfinder

rutt wrote:

Gmonkeh posted a shot on the Whipping Post. Here it is side by side with my edit.



I think it should look more like my edit than the orginal. In fact, unless you are looking for some sort of very special effect, I know it should look more like that. If you print the former, you will surely be unpleasantly surprised.

How do I know? I know because I measured the color values of her skin, especially on her forearms and chest (where there is unlikely to be makeup) and found that they are much more magenta than yellow. Long experience has taught us that nearly 100% of viewers don't like this compared with the opposite except in very special circumstances (sunburn, rosy cheeks, outside in a snowstorm.) Certainly not for beauty. This is so well understood a phenomenon and so often reinforced by new data as to be close to science instead of an artistic choice.

I corrected the orginal with a simple change to the blue curve:



(There are many other ways to do this, but this one is should be a basic part of everyone's toolbox.) What's going on here? Yellow and blue are opponent colors. Yellow pigment doesn't reflect blue and blue light doesn't reflect from yellow surfaces. This curve reduces the blue through the low midtones where the flesh resides. Reducing the blue, comes to the same thing as increasing the yellow, allowing better balance with the magenta.

So much for theory. You don't actually need to know it. Bring up RGB curves, select the blue channel. Hold down the mouse button and mouse over the relevant parts of the image. You will see where they lie on the curve. Read the numbers from the info palette. In LAB, A should be no greater than B. In CMYK, M should be no greater than Y. In RGB, green should be a little higer than blue (same thing as Y being greater than M.)

There is a great tutorial on this by Baldy.

Still with me? I have to say something here that I have said many times before. DON'T TRUST YOUR EYES. DON'T TRUST YOUR CALIBRATED MONITOR. JUST SAY NO. The reason? Your visual system really really wants to correct for ambient light and show you correct skin tones for healthy people. Since this is obviously a healthy young woman, you will tend to see a healthy skin tone here, at least on your monitor which is relatively more immersive than a print. As you are the photographer and have looked at this image quite a bit, this will be even more true. Moral: Always measure the color values of skin tones.

Thank you for posting this John, this is very insightful and I missed it. I do agree that the first image looks too pale, but actually is magenta and missing the needed yellow.

I suggested a method for correcting faces that are too reddish due to being captured by digicams and their flashes in which I suggested applying a green filter to the image, as I contended that the faces were not too red but too magenta. Green kills magenta, and this was a global filter not a selected edit. I see from your comment - in RGB, Green should be a little higher than Blue - that my suggestion may make sense. Using a curve to limit the effect to just the mid tones is even better.

I will bookmark this thread. Thanks again.

rutt

pathfinder wrote:

I see from your comment - in RGB, Green should be a little higher than Blue - that my suggestion may make sense. Using a curve to limit the effect to just the mid tones is even better.

Yeah, just keep saying to yourself "CMYK is RGB".

rutt

pathfinder wrote:

Thank you for posting this John, this is very insightful and I missed it. I do agree that the first image looks too pale, but actually is magenta and missing the needed yellow.

I didn't want to confuse the basic issue, but I thought the picture was still too pale and the face too flat after I fixed the midtone cast. That's only the first step of the Dan Marguls Portrait technique. Just for fun, I pushed it through the rest of the technique:

pathfinder

On my old Sony LCD, it looks like you have lost a just a little detail in her hair along the left side of her face; it certainly has a more dramatic look to it.

pathfinder

rutt wrote:

Yeah, just keep saying to yourself "CMYK is RGB".

SO - why, then, did you do your curve in the blue channel, instead of the yellow channel in CMYK?

I suspect you had reasons for choosing the blue curve rather than the Yellow curve and I am interested in hearing them.

rutt

pathfinder wrote:

SO - why, then, did you do your curve in the blue channel, instead of the yellow channel in CMYK?

I suspect you had reasons for choosing the blue curve rather than the Yellow curve and I am interested in hearing them.

It didn't matter. RGB is CMYK. It was already in RGB. So I saved a trip to CMYK. This little fixup should a standard first step with portraits if they need it and no reason to add extra steps. If I wanted to work on those shadows, I might have got to CMYK for that. And when it's all done, I might take it to CMYK for a little cooling and shadow work. But try blending the green into the red at low opacity for the cooling. And the shadows were already overdone in the photographer's original post.

rutt

pathfinder wrote:

On my old Sony LCD, it looks like you have lost a just a little detail in her hair along the left side of her face; it certainly has a more dramatic look to it.

Might be, but I think the photographer had already plugged these shadows in his original post and so I assume that's the look he is going for.