Color Correction & CMYK

chrisjlee

I've been told by a friend that when you print a photo, that have skin tones, you must ensure that your Cyan values, found on the info pallette of PS CS, should be 1/3 of your magenta value in order for the skin to look somewhat normal. What are the best methods I should use to bump down the magenta values? Also, Take in consideration that if i've been working in LAB mode before converting to the CMYK color profile.

chrisjlee

Nevermind. This solved it: http://www.smugmug.com/help/skin-tone

rutt

You might not need to make the trip to CMYK if your press doesn't demand it. Most inkjets want RGB and do their own conversions. In LAB, you want the A and B values both positive with B >= A. Edgework, who does this for a living, things you should use selective Color to bump the cyan values a touch (maybe 5%) in both the reds and yellows after using the Dan Margulis Chapter 16 technique. I haven't be doing that, but maybe I should. The equivalent in LAB would be to bump the L curve a bit through the flesh tones (make them a bit darker.)

edgework

rutt wrote:

You might not need to make the trip to CMYK if your press doesn't demand it. Most inkjets want RGB and do their own conversions. In LAB, you want the A and B values both positive with B >= A. Edgework, who does this for a living, things you should use selective Color to bump the cyan values a touch (maybe 5%) in both the reds and yellows after using the Dan Margulis Chapter 16 technique. I haven't be doing that, but maybe I should. The equivalent in LAB would be to bump the L curve a bit through the flesh tones (make them a bit darker.)

I've been using the selective color move lately in images that seem to be badly deficient in cyan (mostly digital shots). It's not a true substitute, however, for actual cyan information. In an ordinary scan, there should be enough information in the Cyan/Red channel to actually apply a curve that would bring the balances up to par. A good rule of thumb: hot flesh tones don't necessarily mean too much magenta; usually it's a symptom of not enough cyan information.

If you then have to actually convert the image to CMYK, you're in even worse shape, since Photoshop will compensate for the expected dot gain on press by sucking even more information out of the channels, and, in the case of a deficient Cyan channel, that can result in virtually no information at all in the skin tones that might respond to a curve.

Dan Margulis' Professional Photoshop has an interesting approach. If you are in RGB or Lab and your destination is CMYK, make a copy of the Red channel before your CMYK conversion. This channel contains essentially the full amount of information in the image before some of it is siphoned off into the black channel, and more is pulled back for dot-gain compensation. You will almost always find much better information in the Red channel skin tones than in the comparable Cyan channel. So what I've done, based on the information in Professional Photoshop, is dupe the base layer (in CMYK) and then switch the Red copy for that layer's Cyan plate. The result will usually be too much Cyan elsewhere in the image, but it puts accurate Cyan information into the flesh tones, which can then be balanced with a curve. I also pull the top anchor down to about 70 or 75, since, in theory, the shadow detail from the Red channel has already been transferred to the black plate and you don't want Cyan plugging up the shadows. Sometimes it's necessary to actually mask outt the skin areas, but usually you can bring the values into balance through curves. (A curve can reduce too much information much more effectively than it can increase a weak channel).

If the image is in CMYK to start with, conveting to RGB can often worrk the dot-gain compensation in reverse, providing a Red channel that improves consideraly on the original Cyan.

Baldy

Great post, edgework. clap

chrisjlee

rutt wrote:

You should use selective Color to bump the cyan values a touch (maybe 5%) in both the reds and yellows after using the Dan Margulis Chapter 16 technique.

Hmm i haven't read that far - i'm only on chapter 10. I feel jumping to ch. 16 is cheating.

rutt

chrisjlee wrote:

Hmm i haven't read that far - i'm only on chapter 10. I feel jumping to ch. 16 is cheating.

By Ch 10, you have enough to make that jump. Unlike chapters 8-15, Ch 16 is not a specific technique or tool, but a general workflow for portraits.

BinaryFx

edgework wrote:

I've been using the selective color move lately in images that seem to be badly deficient in cyan (mostly digital shots). It's not a true substitute, however, for actual cyan information.

Agreed Edgework, which is where channel blending can help. For speed I favour a simple channel mixer adjustment layer move, in the red/cyan channel dropping red/cyan to around 80% and mixing in 20% of the green/magenta channel. This will kill all reds though, and affect other hues. This can be restricted via a layer mask refined off the red/cyan channel, with blend if sliders or with a color blend instead of normal blend (and of course opacity reduction).

If one has more time or the image is a hero or cover shot, channel blending via the apply image command and duped layers does similar to channel mixer but offers more control (but is less interactive). Perhaps blend 20% of the green/magenta channel into the red/cyan in darken blend mode in a duped layer. One can also add blend if sliders or layer masks to restrict the edit more to the skintones etc.

http://www.ledet.com/margulis/Numbers.pdf
http://www.ledet.com/margulis/Poetry.pdf

In an ordinary scan, there should be enough information in the Cyan/Red channel to actually apply a curve that would bring the balances up to par. A good rule of thumb: hot flesh tones don't necessarily mean too much magenta; usually it's a symptom of not enough cyan information.

Agreed on the lack of a good cyan or red channel for bad skin reproduction or lack of detail/contrast in faces.

Many digital cameras are sensitive to or seem to enhance red in skintones, over and above the human observer.

It should be noted that one can correct skin and other memory colours direct in RGB with CMYK info palette readings, one does not have to do a mode change to CMYK if one does not need/want to.

If you then have to actually convert the image to CMYK, you're in even worse shape, since Photoshop will compensate for the expected dot gain on press by sucking even more information out of the channels, and, in the case of a deficient Cyan channel, that can result in virtually no information at all in the skin tones that might respond to a curve.

Dan Margulis' Professional Photoshop has an interesting approach. If you are in RGB or Lab and your destination is CMYK, make a copy of the Red channel before your CMYK conversion. This channel contains essentially the full amount of information in the image before some of it is siphoned off into the black channel, and more is pulled back for dot-gain compensation. You will almost always find much better information in the Red channel skin tones than in the comparable Cyan channel. So what I've done, based on the information in Professional Photoshop, is dupe the base layer (in CMYK) and then switch the Red copy for that layer's Cyan plate. The result will usually be too much Cyan elsewhere in the image, but it puts accurate Cyan information into the flesh tones, which can then be balanced with a curve. I also pull the top anchor down to about 70 or 75, since, in theory, the shadow detail from the Red channel has already been transferred to the black plate and you don't want Cyan plugging up the shadows. Sometimes it's necessary to actually mask outt the skin areas, but usually you can bring the values into balance through curves. (A curve can reduce too much information much more effectively than it can increase a weak channel).

If the image is in CMYK to start with, conveting to RGB can often worrk the dot-gain compensation in reverse, providing a Red channel that improves consideraly on the original Cyan.

I would look into the blend if sliders to see if they can offer more speed and or quality to the repacement of the cyan channel with the red.

http://members.ozemail.com.au/~binaryfx/blendif.html

There should be less need for this move with the initial channel mixing or channel blending to improve the red/cyan before conversion to CMYK.


Sincerely,

Stephen Marsh.
http://members.ozemail.com.au/~binaryfx/

edgework

BinaryFx wrote:

Agreed Edgework, which is where channel blending can help. For speed I favour a simple channel mixer adjustment layer move, in the red/cyan channel dropping red/cyan to around 80% and mixing in 20% of the green/magenta channel. This will kill all reds though, and affect other hues. This can be restricted via a layer mask refined off the red/cyan channel, with blend if sliders or with a color blend instead of normal blend (and of course opacity reduction).

If one has more time or the image is a hero or cover shot, channel blending via the apply image command and duped layers does similar to channel mixer but offers more control (but is less interactive). Perhaps blend 20% of the green/magenta channel into the red/cyan in darken blend mode in a duped layer. One can also add blend if sliders or layer masks to restrict the edit more to the skintones etc.

http://www.ledet.com/margulis/Numbers.pdf
http://www.ledet.com/margulis/Poetry.pdf




Agreed on the lack of a good cyan or red channel for bad skin reproduction or lack of detail/contrast in faces.

Many digital cameras are sensitive to or seem to enhance red in skintones, over and above the human observer.

It should be noted that one can correct skin and other memory colours direct in RGB with CMYK info palette readings, one does not have to do a mode change to CMYK if one does not need/want to.



I would look into the blend if sliders to see if they can offer more speed and or quality to the repacement of the cyan channel with the red.

http://members.ozemail.com.au/~binaryfx/blendif.html

There should be less need for this move with the initial channel mixing or channel blending to improve the red/cyan before conversion to CMYK.


Sincerely,

Stephen Marsh.
http://members.ozemail.com.au/~binaryfx/

Channel blending, either with Apply image or a Channel Mixer layer, is a poor substitute for the same reason that Selective color is. It doesn't create new, valid information, but, rather, uses the greyscale profile of another channel as a conduit to squeeze new information into the deficient red/cyan plate. A channel blend of magenta into cyan doesn't create new cyan, but, rather, phoney cyan that mimicks the magenta channel. At it's exteme, you end up with purple. The end result of all channel blends, carried to their extremes, is a duotone, with all plates sharing equally the same information.

That is why I was so attracted to Dan Margulis' red plate substitution. It adds additional, actual, information from that channel into an image that is deficient in it. Copying a red channel into the cyan channel of a CMYK conversion, produces a vastly different image than simply blending green/magenta into the red/cyan channel.

Yes, blend-if sliders can be useful for targeting specific ranges in the blend, but I prefer to use the black channel as a layer mask, keeping the original blacks and leaving only the areas most affected by the three color channels.