Contrast and sturation: a question for people who know more about color theory

[DanK]

One topic that I and a few others have posted about from time to time is that if you adjust tonality with RGB tools, like a curve in Lightroom or in Photoshop with a normal blend mode, you also change saturation. In fact, some time back, Manfred and I had a detailed thread comparing the two options in Photoshop for avoiding this link: working in the LAB color space, and working in an RGB space but using the luminosity blend mode.

I went back to this topic recently because a few of us are considering whiling away our hours of seclusion by offering online mini-seminars on specific postprocessing skills, and this is on the list. (I actually did one in an in-person meeting of a local photo club a year ago.) I realized that I can't explain why this happens. If one imposes a curve in an RGB color space, one is actually imposing curves in all three color channels. However, the curve is changing the distribution of luminosity in each channel.

Can anyone explain this or point me to a source I can read that explains it?

thanks very much.

Dan

[xpatUSA]

One topic that I and a few others have posted about from time to time is that if you adjust tonality with RGB tools, like a curve in Lightroom or in Photoshop with a normal blend mode, you also change saturation.
Can anyone explain this or point me to a source I can read that explains it?

Dan,

Which of the several definitions of "saturation" is being referred to?

If in LCh space, one changes L*, The saturation in HSV changes, according to Lindbloom:

https://en.wikipedia.org/wiki/HSL_and_HSV#Saturation

Based on HSV or HSB saturation Sv = (MaxRGB - MinRGB)/(MaxRGB)

https://en.wikipedia.org/wiki/HSL_an...Hue_and_chroma

In other words, changing the Lightness in L*a*b* while keeping a*b* constant changes the HSV saturation.

[Round Tuit]

One topic that I and a few others have posted about from time to time is that if you adjust tonality with RGB tools, like a curve in Lightroom or in Photoshop with a normal blend mode, you also change saturation. In fact, some time back, Manfred and I had a detailed thread comparing the two options in Photoshop for avoiding this link: working in the LAB color space, and working in an RGB space but using the luminosity blend mode.

I went back to this topic recently because a few of us are considering whiling away our hours of seclusion by offering online mini-seminars on specific postprocessing skills, and this is on the list. (I actually did one in an in-person meeting of a local photo club a year ago.) I realized that I can't explain why this happens. If one imposes a curve in an RGB color space, one is actually imposing curves in all three color channels. However, the curve is changing the distribution of luminosity in each channel.

Can anyone explain this or point me to a source I can read that explains it?

thanks very much.

Dan

I would think that the reason is that in "normal mode" you change all three channels by the same amount thus changing their relative proportions where as in "luminosity mode " you change the channels in different proportions(i.e. .2 red,.7 green and .1 blue) which may maintain the same saturation.
Ted probably as the explanation that you are looking for

[xpatUSA]

I would think that the reason is that in "normal mode" you change all three channels by the same amount thus changing their relative proportions where as in "luminosity mode " you change the channels in different proportions(i.e. .2 red,.7 green and .1 blue) which may maintain the same saturation.
Ted probably as the explanation that you are looking for

Indeed, André,

The only way saturation (HSV) could NOT change is if the max and min RGB values remain the same!

[DanK]

Thanks very much. To make my question clearer: if I simply pulled the curve up, thus increasing R, G, and B values, it would intuitive sense to me. But imposing a curve needn't increase mean values at all. It can simply redistribute values, stretching out the midtones.

I'll try to put together an example. Oddly enough, I have too much to do today, so I am not sure I'll get to it, but I'll post if I do.

[xpatUSA]

Thanks very much. To make my question clearer: if I simply pulled the curve up, thus increasing R, G, and B values, it would [make] intuitive sense to me. But imposing a curve needn't increase mean values at all. It can simply redistribute values, stretching out the midtones.

Hmm ... Not being familiar with Adobe stuff, some numbers or curve settings would help. Meanwhile the Wiki link should avoid the need for playing around with sliders to see wo hoppen.

I'll try to put together an example.

If we knew which of the several definitions of "saturation" you are referring to, perhaps we could help more.

[DanK]

[QUOTE=xpatUSA;751223]Hmm ... Not being familiar with Adobe stuff, some numbers or curve settings would help. Meanwhile the Wiki link should avoid the need for playing around with sliders to see wo hoppen.

I'm honestly not certain at this point, but I think the ratio of colorfulness to brightness, as I am only interested in perceived saturation.

I have to prepare for a class and a dissertation defense, even though I am retired, so I may not get to this soon. However, if I can do it, I will include a screen shot of the curve, along with RGB and LAB values for specific locations in the image so that none of it is Adobe-reliant.

[xpatUSA]

Definition of saturation

I'm honestly not certain at this point, but I think the ratio of colorfulness to brightness, as I am only interested in perceived saturation.

From another Wiki:

"Saturation is the proportion of pure chromatic color in the total color sensation.[11]



where Sab is the saturation, L* the lightness and C*ab is the chroma of the color."

Pardon the black on gray formula.

https://en.wikipedia.org/wiki/Colorfulness

You can get L* and C*ab for a given RGB from Lindbloom's calculator ...

[xpatUSA]

P.S. In RawTherapee's RGB Curves, the HSV saturation also only changes a bit when the Luminosity box is checked. Sound familiar?

Perhaps that is because Luminosity is similar to L* or even equal in Adobe-stuff, ¿ quien sabe ?

So, if the Chroma stays the same and the Luminance is held constant, it makes sense that the "saturation" won't change much, if at all.

[xpatUSA]

P.P.S. It occurs to me that if an 'S' curve is applied to any or all of RGB while keeping the mid-point fixed, only mid-tone pixels will retain the same saturation. All others will change UNLESS that Luminance box is checked. Not saying anything new - it was just a passing thought.

[DanK]

P.P.S. It occurs to me that if an 'S' curve is applied to any or all of RGB while keeping the mid-point fixed, only mid-tone pixels will retain the same saturation. All others will change UNLESS that Luminance box is checked. Not saying anything new - it was just a passing thought.

Indeed. I have been playing hooky from what I should be doing and playing with some contrast adjustments, using both rgb (normal blend) and lab (luminosity) adjustments. I think what is happening is a perceptual issue. Just as we are more sensitive to luminosity differences in the midtones, I suspect we are more sensitive to R, G, and B differences in the midtones. Applying a curve anchored at the midpoint is simply spreading out the range of R, G, and B values over most of the distribution, while making room for that by compressing the ranges in the tails.

I haven't taken the time to redo this with specific values, but just to show the effects, the two images below, which I created a long time ago, have the same extreme contrast adjustment via a curve, the first with a normal blend and the second with a luminosity blend.

[xpatUSA]

I haven't taken the time to redo this with specific values, but just to show the effects, the two images below, which I created a long time ago, have the same extreme contrast adjustment via a curve, the first with a normal blend and the second with a luminosity blend.

Interesting. I'm unfamiliar with blend modes but they are certainly different. Can't deduce much at glance as the reds are blown in both and the blues are bottomed in one (RawTherapee histogram).

[xpatUSA]

Interesting. I'm unfamiliar with blend modes but they are certainly different.

While reading about blend modes just now, I found this, FYI:

"In photo editing, changing the blend mode of a layer to Luminosity is often a final step. For example, a very common photo editing technique is to use either a Levels or Curves adjustment layer to improve overall contrast in an image, and in many cases, this works perfectly. The problem you can run into, though, is that Levels and Curves affect not only the lightness values in an image, they also affect color. By increasing image contrast, you're also increasing color saturation, especially in reds and blues, and sometimes you'll even see a shift in colors. Too much color saturation in a photo can wipe out important image details. By changing the Levels or Curves layer to the Luminosity blend mode, we easily avoid the problem by telling Photoshop to ignore the color information completely." (My emph).

https://www.photoshopessentials.com/...es/luminosity/
.

[DanK]

I can explain the blend modes. "normal" is what you would get in almost any software applying a curve in an RGB color space--which is to say, in almost any software if you don't take special steps not to do this. Using a luminosity blend mode is almost identical to working in LAB mode and applying the curve only to the L channel. (Manfred and I had a long discussion of this some time ago. To test this, I did identical adjustments both ways. The results were essentially the same. There were only trivial differences, which I assumed were just minor differences in algorithms.)

I don't have anything left from those other than the output JPEGs, but I believe the clipping is there only because I deliberately made the contrast adjustment extreme to make the difference in output extreme. You get the same qualitative difference with a more modest adjustment that doesn't clip, just less extreme.

[xpatUSA]

I can explain the blend modes. "normal" is what you would get in almost any software applying a curve in an RGB color space--which is to say, in almost any software if you don't take special steps not to do this. Using a luminosity blend mode is almost identical to working in LAB mode and applying the curve only to the L channel. (Manfred and I had a long discussion of this some time ago. To test this, I did identical adjustments both ways. The results were essentially the same. There were only trivial differences, which I assumed were just minor differences in algorithms.)

Thanks - I'm reading Blending 101 even as we speak.

I don't have anything left from those other than the output JPEGs, but I believe the clipping is there only because I deliberately made the contrast adjustment extreme to make the difference in output extreme. You get the same qualitative difference with a more modest adjustment that doesn't clip, just less extreme.

Yes, I imagined that would be the case.

[Manfred M]

I think you are definitely figuring this out Dan.

Yes, human vision is most sensitive in the mid-tones and I seem to remember seeing an image of a modified "S" curve that we have less sensitivity to colour differences in the shadow details and in the highlights with maximum sensitivity in the mid-tones.

When using the normal blend mode, changes to any and all of the R, G and B colour channels inherently means you are also impacting the saturation. We tend to notice this more in the deeper tones than the lighter ones. By working with the L* channel, one deals only with lightness values, so changing the curve here lets the user light or darken specific areas of the image without changing the saturation at all.

What I've always found interesting when working in the L*a* b* colour space in Photoshop is how broad the distribution of values is in the L* channel and how narrow the range is in the a* and b* channels.

[DanK]

Manfred,

One reason reason I generally use a luminosity blend in RGB space rather than LAB mode is just familiarity: I find it harder to work with color changes on the a and b axes. One can accomplish much the same thing by using the color blend mode while in an RGB space. I think if I used LAB enough, that disadvantage would decrease over time.

Dan

[Manfred M]

Manfred,

One reason reason I generally use a luminosity blend in RGB space rather than LAB mode is just familiarity: I find it harder to work with color changes on the a and b axes. One can accomplish much the same thing by using the color blend mode while in an RGB space. I think if I used LAB enough, that disadvantage would decrease over time.

Dan

Dan - I absolutely agree and colour changes in Lab colour space are non-intuitive to state it mildly. I generally ensure that I get my colours looking right in the raw converter (ACR is my usual choice) and after that I only change the lightness value (which can also change contrast).

The other issue when using Lab is that some of the other functions / filters and especially plug-ins do not work.