ezpkns retouching wrote:
just pulled a little magenta out of the Blues - then a little bit more, then a little bit more, and then I started to really doubt its usefulness as the luminosity and saturation of the fabric shifted far too much.

Theres no problem with using Selective Colour to desarurate colours, in fact its better to do it this way than with Hue/sat:

Its adding yellow to blues and to cyans that will desaturate them. Try this in Selective Colour.  The result (not so destructive) will look much the same as if you had pulled saturation out of blues in Hue/sat. Luminosity shifts can be controlled with "blacks" which infact  puts neutral density into the image (while in RGB).

Saturation can be controlled with any colour adjustment. Hue Saturation is actually an very destructive adjustment, because when using 'hue' it moves channel data around, crudely masked from surrounding areas. Of course Curves can also be used to carefully put the 'unwanted' colour into highly saturated colors. "Unwanted' means the opposing color.

It is what you put into saturated colours that desaturates them. Find the opposite colour and introduce that. You might think logic would tell you that its what you take out, but thats not how it works.

Robert Randall wrote:
The HS and Selective Color tools are approximately the same thing,

Unfortunately that is untrue - they use completely different mathematical means, Hue saturation does create harsh transitions and is more destructive- its the Hue/sat hue hue slider and channel mixer that are interchangeable. Even Curves and Levels (endpoints and midpoints i mean for anyone determined to dispute me) even though they produce similar visual result actually use different math so are not completely interchangeable.

Robert Randall wrote:
Finally, its not the color that you put into a color that desaturates it, it is the density you pull out that desaturates it.

Again not the whole case I'm afraid, suggesting someone who has never worked in press. A single colour or combination of colours can be added to any saturated colour to desaturate it. Yellow to blue cyan to red. Much less problematic in the world of CMYK than taking ink out. Pulling ink out on press will affect both luminance making colors washed out. 

Read Dan Margulis very old book on colour retouching - he says precisely the same thing

Your point about not exceeding total density is of course true, but in practice slightly boosting most colour channels will not push density further than max ink. Density is controlled in CMYK partly with the black channel

Snap2 wrote:
Unfortunately that is untrue - they use completely different mathematical means, Hue saturation does create harsh transitions and is more destructive- its the Hue/sat hue hue slider and channel mixer that are interchangeable. Even Curves and Levels (endpoints and midpoints i mean for anyone determined to dispute me) even though they produce similar visual result actually use different math so are not completely interchangeable.

Can you provide an Adobe or Adobe-based source for this information?  First I've heard of it.

Snap2 wrote:
With respect most of that post seems to have been designed to dispute what I carefully wrote in mine

Robert Randall wrote:
The HS and Selective Color tools are approximately the same thing,

Unfortunately that is untrue - they use completely different mathematical means, Hue saturation does create harsh transitions and is more destructive- its the Hue/sat hue hue slider and channel mixer that are interchangeable. Even Curves and Levels (endpoints and midpoints i mean for anyone determined to dispute me) even though they produce similar visual result actually use different math so are not completely interchangeable.


Again not the whole case I'm afraid, suggesting someone who has never worked in press. A single colour or combination of colours can be added to any saturated colour to desaturate it. Yellow to blue cyan to red. Much less problematic in the world of CMYK than taking ink out. Pulling ink out on press will affect both luminance making colors washed out. 

Read Dan Margulis very old book on color retouching - he says precisely the same thing

Your point about not exceeding total density is of course true, but in practice slightly boosting most colour channels will not push density further than max ink. Density is controlled in CMYK partly with the black channel

I did not write anything to dispute your efforts, I wrote it because it is true and factual. I have no intention of getting into a pissing match with anyone, but after 35 years of prepress work, I'm not going to stand idly by when I see situations that can be better served by a different workflow.

If you have a magenta screen plate that is 100% density, how does adding yellow or cyan to the file, decrease the saturation of the magenta plate. You will be raising the gray component, and possibly the appearance of saturation will diminish, but the magenta plate will remain at 100%, which is completely saturated.

I have spent somewhere in the neighborhood of a couple hundred hours in the same room as Dan Margulis, hanging on every word he spoke, and I have probably read the book you're referencing at least a dozen times if not more. What he says about the unwanted color is that you can use it to retain or increase shape in an object. Hence the example picture of the lime in his color theory book. While adding unwanted color to a file may actually gray out the affected areas and give the appearance of desaturation, you are still left with a problem of ink densities.

I never suggested removing ink on press was an option, I simply said the pressman has ink levels as a control device should he want it, and typically, no file makes its way to final ink on paper without the pressman exerting his control over ink densities. And sometimes they do pull color out.

Selective color and hue/saturation tools may or may not use the same math to get to an end point, I couldn't tell you. What I've been told, and shown the graphs to back it up, is that they both use a matrix to arrive at their conclusion, and that is all I said. In my thinking, that makes them interchangeable.

Typically, contrast is controlled by the black channel, and density is controlled by all channels in concert. Although you can elect to control density with the black channel, it will result in a loss of chromaticity. This is well illustrated in the resultant file when you increase the black component of a GCR separation.

You mentioned a number of times the math component in the tools we're discussing. Since I have no idea what those equations are, and you seem to know, can you provide the source for them. Also, everyone I've ever spoken with about the difference between curves and levels says they are identical until you start introducing movement in the curve between the endpoints and the mid point.

Snap2 wrote:
Ok

Hue/saturation evens the channels - in order to do this it has to keep luminosity the same, so it does its own very crude form of balancing act.

Taking saturation down in a CMYK file using Hue/sat will make it brown not gray, try taking it to zero as an experiment. This is because it evens out the three colour channels, making brown not neutral gray. That's the first reason that using hue/sat in CMYK is not advised and never has been.

It doesn't make anything brown, it makes it perfectly neutral, which in the CMYK world, means it will proof a slightly reddish cast due to the impurities in the cyan ink. The brown you reference is Adobe's attempt at approximating that cast through a soft proof on your monitor. Take that same file to a pre press shop and have it proofed on a Kodak Approval, and it will simply appear to have a very slight red cast. Adobe's soft proof approximation of that cast is completely off target.

Robert Randall wrote:
If you have a magenta screen plate that is 100% density,

No one ever said anything aboit 100% density, you did

Robert Randall wrote:
What I've been told, and shown the graphs to back it up, is that they both use a matrix to arrive at their conclusion

I have absiluetly no idea what the 'matrix' is apart from a very well made movie starring Keanu Reeves. Selective color uses the smoothest transitions the drop off between color groups is better and smoother. Basic experimentation will show you this

Very few people in professional colour correction use hue/sat - and its actually something that people are trained not to do, in CMYK

Robert Randall wrote:
Typically, contrast is controlled by the black channel, and density is controlled by all channels in concert. Although you can elect to control density with the black channel, it will result in a loss of chromaticity.

Contrast can be controlled in a variety of ways in CMYK. The black channel alone will not do a sufficient job in a colour file. Monochome or undersaturated images benefit from a stronger black channel, and in these rare files yes the black channel can be used successfully to control contrast, but its ridiculous to suggest that that's the way to do it. "after 35 years of prepress work" yeah well you are demonstrating some curious lack of basic knowledge I have to say Robert. Sorry

Snap2 wrote:
I have absiluetly no idea what the 'matrix' is apart from a very well made movie starring Keanu Reeves. Selective color uses the smoothest transitions the drop off between color groups is better and smoother. Basic experimentation will show you this

You know the inner code for PS, but don't know what a color matrix operation is?  Funny how education works.

Still looking for that Adobe citation?

Robert Randall wrote:
It doesn't make anything brown, it makes it perfectly neutral, which in the CMYK world, means it will proof a slightly reddish cast due to the impurities in the cyan ink. The brown you reference is Adobe's attempt at approximating that cast through a soft proof on your monitor. Take that same file to a pre press shop and have it proofed on a Kodak Approval, and it will simply appear to have a very slight red cast. Adobe's soft proof approximation of that cast is completely off target.

Oh ferchrissake Robert, take a CMYK file and TRY IT. Use the eyedropper measure your channels see that C.M and Y are the same that equals Brown not gray. Impurities in the cyan ink are what makes it brown. The word neutral means gray. If you can't see that's brown, then you need to recalibrate your monitor.

Snap2 wrote:

Robert Randall wrote:
If you have a magenta screen plate that is 100% density,

No one ever said anything aboit 100% density, you did

Robert Randall wrote:
What I've been told, and shown the graphs to back it up, is that they both use a matrix to arrive at their conclusion

I have absiluetly no idea what the 'matrix' is apart from a very well made movie starring Keanu Reeves. Selective color uses the smoothest transitions the drop off between color groups is better and smoother. Basic experimentation will show you this

Very few people in professional colour correction use hue/sat - and its actually something that people are trained not to do, in CMYK


Contrast can be controlled in a variety of ways in CMYK. The black channel alone will not do a sufficient job in a colour file. Monochome or undersaturated images benefit from a stronger black channel, and in these rare files yes the black channel can be used successfully to control contrast, but its ridiculous to suggest that that's the way to do it. "after 35 years of prepress work" yeah well you are demonstrating some curious lack of basic knowledge I have to say Robert. Sorry

You are behaving like an asshole, which means you are no longer worth my time.

Snap2 wrote:
Did you even bother to do my tests? Did you manage to create a levels and curves that did exactly the same thing?

I was actually more curious as to what you were basing the Hue / Sat assertions on?   I'll try to the curves / levels thing after bit.

Snap2 wrote:
Ok

Hue/saturation evens the channels - in order to do this it has to keep luminosity the same, so it does its own very crude form of balancing act.

Taking saturation down in a CMYK file using Hue/sat will make it brown not gray, try taking it to zero as an experiment. This is because it evens out the three colour channels, making brown not neutral gray. That's the first reason that using hue/sat in CMYK is not advised and never has been.

If you mean about the Curves and Levels - you will not be able to produce the exact same result using the midpoint value in curves as you will with the midpoint slider in levels- use the difference method , or Apply Image subract 1 offset 128, and do two adjustments, and you will not be able to precisely create the same pixel values

Until just now, I've not used Hue/Sat in CMYK, as I rarely work in CMYK, and besides, I was told not to. It is interesting, regardless of the underlying mechanics, but I find the differences in behavior of all the color modes fascinating.

The Curves/Levels difference is interesting as well. Just because I have no current use for the information, doesn't mean it'll never come in handy. Thanks for contributing.

Snap2, meet Robert Randall. Not one to toot his own horn, let me just say that he's something of an "elder statesman" here, and also very generous about sharing his decades of experience in photography, prepress and digital imaging.

EDIT: got distracted while typing this reply - apparently in the interim you have gotten better acquainted already.

ezpkns wrote:
Snap2, meet Robert Randall. Not one to toot his own horn, let me just say that he's something of an "elder statesman" here, and also very generous about sharing his decades of experience in photography, prepress and digital imaging.

Yes I realise this. But it doesn't stop someone coming in who happens to know more about a single subject does it? really should this be the case? no-one, myself included, can pretend to be an expert at everything. I spent 4 years colour correcting every day and night, I'm afraid that entitles me to some knowledge. And I'm absolutely sure Robert knows many things that I don't.

ezpkns wrote:
The Curves/Levels difference is interesting as well. Just because I have no current use for the information, doesn't mean it'll never come in handy. Thanks for contributing.

Yes I dont think many people would have much use for that information, point taken

Snap2 wrote:
Sean to defend my hue/sat argument - Take a hue sat layer and move the Hue slider in an RGB file. Now look at your channels. What you are seeing is the Red green and blue channels directly SWAPPING with each other. Exactly the same as a Channel mix. In CMYK thats like trying to do an operation with a pickaxe

I understand what you're saying about the H/S layer in that context; my only argument is that there is no 'proof' that these are so different.  Certainly, it doesn't respond well in CMYK, but that doesn't actually need to be so.  Running it through a color matrix can allow conversion to, operation on (adjusting hue), and return to CMYK in a few simple operations.  The limiting factors become the bit depth worked at originally (8bit users will always suffer more here) and the accuracy to which the internal constants operate within PS itself (something everyone will suffer).

As to the levels / curves thing, I actually am going to have to disagree with you here, at least as far as capacity is concerned.  I spent about ~1-2 minutes to make a pseudo-gamma curve which replicated a levels midpoint adjustment of 1.5 to within 2/255 accuracy in each channel.  I feel rather confident that a bit more effort at refinement could do even better, though I'm not inclined to spend the time at the moment.  For its purpose, levels would be an easier option, but one shouldn't interpret that to mean that curves isn't capable of producing the same result, as it clearly can.

Sean Baker wrote:
but one shouldn't interpret that to mean that curves isn't capable of producing the same result,

The same mathematical result no, visually yes - do an analysis of the MATH difference using the Apply Image I suggested. In this method, used to accurately determine the differences in two layers, 128 gray equals precisely the same value anything else doesnt.

As I still have no idea what a 'matrix' is, it would be useful to know why the Selective Color produces such lovely smooth transitions between colour groups, and Hue sat looks like its been carved with a broad sword.

Snap2 wrote:
The same mathematical result no, visually yes - do an analysis of the MATH difference using the Apply Image I suggested. In this method, used to accurately determine the differences in two layers, 128 gray equals precisely the same value anything else doesnt.

I know how to do it, and that's what I said - with 90 seconds effort, I had a curve which was accurate to 2/255 in each channel.  By your technique, that would result in values with a maximum of 127 or 129 in the gray layer.  Not 100% accurate, but not a lot of time spent.   I'm not arguing that it's easy to make such a curve (though one could easily write a program to output a desired gamma curve to a PS curve file), only that it is possible without causing destruction of the image data.  It's no big deal - levels is still more efficient to the singular task, only mathematically intriguing that they're still capable of doing the same things in different ways.  Figuring out how to pull down the Curves midpoint in Levels would be a lot more amusing IMO .

Snap2 wrote:
As I still have no idea what a 'matrix' is, it would be useful to know why the Selective Color produces such lovely smooth transitions between colour groups, and Hue sat looks like its been carved with a broad sword.

Sadly, this may be as simple as how old the code is within PS, and whether anyone updated the HS code when they introduced adjustment layers and brought Selective Color on board.  Clearly they use different math; it's simply a question of with what precision that math is being performed and what the accuracy of the operands in each is.

More generally, a color matrix is simply a mathematical matrix against which color values from one color space are multiplied in order to produce a second matrix containing the values of a different color space.  Hence why precision both of the variables themselves as well as the constants (values within the matrix - perhaps your culprit in your larger questions) is so vital.

ezpkns wrote:
The Curves/Levels difference is interesting as well. Just because I have no current use for the information, doesn't mean it'll never come in handy. Thanks for contributing.

Snap2 wrote:
Yes I dont think many people would have much use for that information, point taken

Actually, that time I wasn't being facetious. I do appreciate everyone's contributions, for as you pointed out, everyone knows different things. And I'm a sucker for weird PS factoids like the Levels Curves thing.

Even desaturating  with Hue/Sat in CMYK... it certainly wasn't neutral, but it wasn't unpleasant. I might never have tried it without this thread. I file it all away for future reference.

Snap2 wrote:
Ok Sean, so the 'matrix' is a way of translating RGB to HSB values mode for a Hue/sat adjustment. And for CMYK to RGB Selective Colour- requires its own matrix? I get it.

Guess thats the reason then that Hue/sat is so much more destructive. CMY is easier to translate into RGB obviously

We are getting there.

You're on track with the matrix piece, though to be accurate, it's widely believed (and supported by insiders and evidence) that all color conversions are twofold within PS - one to its internal LAB or LAB-based representation, and another to to the target.  I.e. RGB -> CMYK is really RGB -> LAB -> CMYK.

CMYK shouldn't necessarily be a lossy conversion between it and HSB, though, as I'm not sure that there are colorspace issues between them.  I'll take a look unless someone more knowledgeable chimes in shortly.

Snap2 wrote:
No of course I'm not suggesting that. Thanks for trying to catch me out. As far as I know it just does the same as it does in an RGB file using the CMY channels

People have long argued that hue/sat should be updated so that it handles true saturation properly in CMYK

It's saturation which is a problem then?  I know you've identified an issue - I just want to replicate it for my own edification .

Snap2 wrote:
No hue is a problem too. Shift the hue in REDS and look what it does to the crossover areas. Selective Colour will eventually do this but its a damn site less destructive. My point all along

Ahhhh, the issue then isn't the math going on way behind the scenes but rather the manner in which the adjustment range is being determined.  Much like Levels is more appropriate to adjusting gamma, then, as a simple workflow issue, Selective Color makes more sense while working in CMYK?

Sean Baker wrote:
but rather the manner in which the adjustment range is being determined.

Yes its not just the range, that can be controlled in Hue/sat with the sliders. Its the method of assessing that range, whatever that is.

Sean Baker wrote:
Selective Color makes more sense while working in CMYK?

Yes and produces a more controllable result. It also works better in RGB than hue/sat does.

Snap2 wrote:

Yes its not just the range, that can be controlled in Hue/sat with the sliders. Its the method of assessing that range, whatever that is.

I suspect, based on the interface and human nature, that it uses the same ugly-*** code which underlies the Blend-If sliders in the layers dialog.  *shudder*

Snap2 wrote:
Yes and produces a more controllable result. It also works better in RGB than hue/sat does.

For Hue adjustments you mean?

Snap2 wrote:
Blend-if uses a version of luminosity, probably a different one to the thing used in gradient mapping and Luminosity blend mode. I dont know. Can't really be the same as the colour ranges in Hue/sat though

I mean in the algorithm it uses to blend the selected region.  That is, I can select a luminosity range with Blend If which I could also select with a couple of layer masks, but the layer masks given me a much better & smoother result.  Similarly, I'm suggesting that the Hue Sat layer uses a very lossy manner of handling the selection range, quite possibly sharing code with the Blend-Ifs.  That the data are in a different format is in many ways irrelevant.

Also fwiw (not much IMO - I think it's a great idea with terribly implementation), Blend-If can base off of any channel, so R, G, B, and gray (close, but not quite Lum) are options while working in RGB.