Effect of Luminosity Blending Mode on Curve Adjustment Layers in Photoshop

[Round Tuit]

I have been working on restoring 55 years old photos that had been taken with point and shoot cameras that most likely did not even have a flash. Needless to say that the prints have not only faded but their colours are now either mostly magenta or green with the odd pink one.

I knew that using curve adjustment layers in normal blending mode to darken faded colour also affected the hue and saturation of the colour. What I did not realise was that the luminosity mode protects the hue but increases the saturation as well as decreasing the brightness of the colour. This is also true using the hue/saturation adjustment layer; that is to say that its hue slider has no effect if the blending mode is luminosity.

This came as a surprise to me as I had the impression that the luminosity mode prevented the increase in saturation when burning colours.

[Manfred M]

There is something strange happening there André.

In an RGB colour space, in normal blending mode saturation increases when luminosity decreases. With luminosity blending mode, only the luminosity should change and there should be not change to saturation.

My normal workflow in the situation you are retouching is to use a Hue / Saturation adjustment layer. Rather than using the "Master" setting I will work with (in your case) the green or magenta slider and the colour cast. I then go back to a curves adjustment layer and work both the luminosity and contrast (two different layers).

If you want to be a bit tricky, convert to the L*a*b* colour space; the L* will only change the luminosity without impacting the saturation or hue (those are handled with the a* or b* channels, but I find them a bit finicky to use).

[Round Tuit]

...In an RGB colour space, in normal blending mode saturation increases when luminosity decreases. With luminosity blending mode, only the luminosity should change and there should be not change to saturation.

That is what I thought should happen but it does not work that way. I am getting ready to go away for the week end. When I get back next week, I will describe the experiment that I did to demonstrate that in luminosity blending mode, the hue stays constant but the saturation increases when the brightness is brought down.

[DanK]

I agree with Manfred. My experience is that the luminosity blend mode affects only brightness. I'll be interested in seeing your test. How are you measuring saturation?

I believe that if you maintain exactly the same saturation but darken the image (luminosity only), the colors will appear more saturated but aren't.

[Round Tuit]

The experiment that I did is as follows:
1) Open a new document in photoshop.
2) Fill the document with a relatively bright but lightly saturated colour using the colour picker tool.
3) Open the "info" panel and set the primary display to RGB and the secondary display to HSB.
4) Add a curve layer in normal blending mode and drag the middle of the curve downward.
5) The info panel will display the before and after values for the RGB and HSB. The after value will show an increase in saturation and a decrease in brightness. The hue will also change if the original value is not a multiple of 60.
6) Change the blending mode of the curve layer to "luminosity". The saturation will increase but not as much as in normal mode. The brightness will decrease less than in normal mode and the hue will not change regardless of the value of the original hue.

This indicate to me that the luminosity mode does not affect only the brightness but rather it keeps the hue constant and lets saturation and brightness vary.

[DanK]

I'm not an expert on color science, but I think I can explain.

If you look at explanations of Photoshop blending modes, including Adobe's own (https://helpx.adobe.com/photoshop/us...ing-modes.html), you'll see that they all say that the luminosity blend mode affects brightness, not color. However, look closely at Adobe's language:

Luminosity
Creates a result color with the hue and saturation of the base color and the luminance of the blend color. This mode creates the inverse effect of Color mode.

So, the original color is unaffected but a layer is added with lower luminance.

If you look up how saturation is calculated, you'll find a bunch of different formulas, but they all entail some variation on dividing chroma (colorfulness) by luminance. so, if you reduce luminance but make no change to color, calculated saturation necessarily increases.

So, I think what's happening is that the numerical quantity called saturation is in fact increasing, but what we mean by this--which is the colorfulness of the image--isn't.

Here are four images to address this:

An unedited base image:



The same image with a severe curve, using a normal blend:



With the same curve, but a luminosity blend:



And finally, the same curve, but using the color blend mode:



You can see very clearly: the luminosity blend darkens shadows and brightens highlights without increasing colorfulness, leading to a washed out appearance.

In another set of test images that I posed years ago, I did the same edits in Lab mode, changing only luminance, and in RGB mode with a luminosity blend. The results were virtually indistinguishable.

[Round Tuit]

Thank you Dan.

I think that we can all agree that looking at an actual photo is the best way to subjectively evaluate the effect of an adjustment layer. This approach however looks at the cumulative effect of several variables simultanously. My aim was to try to understand how a curve layer works both in normal and luminance mode. Hence, I picked a single colour and modified it using a darkening curve. I repeated the experiment using different colours with brightening as well as darkening curves.

I found the result of my experiment surprising. In normal mode, a darkening curve increases saturation, decreases brightness and ,depending on the colour, changes the hue. I did not expect that some colours did not shift while other can shift by as much as 9 degrees. In luminance mode, the changes in saturation and brightness are less pronounced and the hue never change. A brightening curve does the opposite. These results are mathematical values where hue goes from zero to 360 degree, saturation is the ratio of the difference between the brightest channel minus the dimmest one over the brightest channel and brightness is the ratio of the brightest channel over the maximum brightness. These formulas are used by Adobe in Photoshop.

Our vision however responds to luminance which is a weighted sum of the three colour channels and therefore is affected by both the brightness and the saturation of the colour.

In conclusion, I will keep using curve adjustment layers in luminance mode and judge the results subjectively. My engineering brain however is happier knowing what goes on under the hood!

[DanK]

and ,depending on the colour, changes the hue

Does this reflect the unbalanced mix of R G and B--that is, the more similar the initial values are, the hue changes less?

In luminance mode, the changes in saturation and brightness are less pronounced and the hue never change.

The last of these is exactly what should happen, as that blend mode is not supposed to act on color.

As a practical matter, what I have found subjectively is that a luminosity blend mode for tonality adjustments sometimes produces an unnatural appearance, while in other cases, a normal blend does. This is mostly an issue with highly saturated colors or in images that need severe tonality adjustments. In many cases, I use some of both. One way I have sometimes done this is to create one adjustment in RGB mode with the severity of the adjustment I want, then duplicate that and set the duplicate to luminosity blend mode, and then reduce the opacity of both to mix them.

[Round Tuit]

Does this reflect the unbalanced mix of R G and B--that is, the more similar the initial values are, the hue changes less?

No. The hue that have at least one channel with a zero value (red, yellow, green, cyan, blue and magenta) do not change. Those hue have a value that is a multiple of 60. The hues with value 30,90,150,210,270,330 change the most. The amount of change ramps up from 0 to max and down to 0 as you go from zero change to zero change around the circle.

[DanK]

No. The hue that have at least one channel with a zero value (red, yellow, green, cyan, blue and magenta) do not change. Those hue have a value that is a multiple of 60. The hues with value 30,90,150,210,270,330 change the most. The amount of change ramps up from 0 to max and down to 0 as you go around the circle.

Thanks. Very interesting.

[The amateur]

As I am now also involved in image retrieval from films, this is another thread where the importance of understanding curves and blending modes becomes clear.
This importance is not only in image retrieval as demonstrated here, but also in its general improvement.

Although, as we know, such mastery requires some initial investment, the excellent final results are magnificent.

Thank you very much to the participants for this contribution to the continued quality of this forum.

[Manfred M]

Interesting André.

I tried a couple of things this afternoon. I dropped sample points onto an image and set the colour reporting to RBG, HSB and L*a*b*.

I did a curves adjustment layer in RGB and when I moved the curve parameters in both Normal and Luminosity mode, the RGB and HSB values changed, but only the L* reading changed, based on the adjustments I made. Something does not add up.

I then converted the image to the L*a*b* colour space and tried both Normal and Luminosity blending modes. As before, the L* channel changed but the RGB and HSL reporting changed, regardless of the blending mode.

I wonder if the issue is that the HSV calculation is not correct, as the L*a*b* values worked as I would have expected.

[DanK]

I just tried this, and I got somewhat different results, but I'm puzzled.

The RGB values should change when luminosity is changed; the RGB values represent the brightness of those channels. This should be true in both Lab and RGB space. However, the issue is the relationships between them. The don't always change uniformly. I have a hunch that the relationships between hue and the RGB values is nonlinear.

As discussed earlier, saturation will change because it's calculated (to oversimplify) as the ratio of chroma to luminance.

What I found after fiddling with a curve and both blending modes is that the hue values fluctuated only a very small amount. Probably calculation error.

So here's my guess about what's happening:

--changing luminance should change either the a and b in lab or hue, although it does seem to change hue slightly in my tests. It should change saturation, depending on whether luminance is increased or decreased. B or V obviously changes. R G and B have to change although not necessarily by the same amount because of scaling issues.

--using a normal blend should have qualitatively but not necessarily quantitatively similar effects. However, the effects on R G and B are going to be different in bright and dark areas, if you use an S-curve. This last I think is the key, but I haven't been able to test it yet because the test images I picked were too high-frequency, and moving the mouse to change the blending mode made it difficult to find the same spot to sample.

[Manfred M]

Not quite Dan.

The L* in the L*a*b* colour space is the lightness channel, so it will only change that value; black has a value of 0 and white a value of 100. A lot of professional retouchers convert their images to this colour space and work only with the L* channel for dodging and burning because it does not change hue or saturation.

The a* and b* change both the hue and saturation. a* is green - red and b* is blue - yellow. The L*a*b* colour space is non-linear and attempts to mimic the response of the human eye. If I remember correctly, my instructor referred to it as a "physiological" colour space due to the way it performs.

The main issue of this colour space is that working with the a* and b* channels is not intuitive and a number of the functions in Photoshop are disabled for it. If you look at the histogram of either of these two channels, it is quite narrow. A lot of plug-ins do not work in this colour space either.

This is a full L*a*b* histogram. The L* channel looks a lot like we see when we look at RGB histograms, the the a* and b* channels are clustered in the mid-area.

[DanK]

Manfred,

I don't think that's right.

I understand why people use Lab, and you'll probably recall that I posted images here a long time ago showing that tonality adjustments using L and the same adjustments using RGB mode and a luminosity blend mode produced essentially identical results.

The RGB values are brightness of those channels, not chroma. You can see this if you pull down midtones with a curve and switch to the color blend mode. The RGB values will remain nearly unchanged. Now, change the same curve to a normal blend mode or a luminosity blend mode. In both cases, the values of the RGB channels will drop because all three colors have been darkened.

The saturation value tabled by Photoshop, as discussed above, is not the same as perceived saturation, that is, perceived color intensity. It's calculated in various ways, but basically as a ratio of chroma to luminance, so darkening an image with L mode or with a luminosity blend will necessarily increase calculated saturation.

I don't know enough color science to understand the relationships among these quantities. However, in practical terms, it seems clear. if you want to change tonality without changing the perceived intensity of color, use Lab mode or RGB mode with a luminosity blend mode. Conversely, if you want to affect the perceived intensity of color without changing luminance, use RGB mode with a color blend mode.

We need the guy from Texas who used to post here, whose name I have forgotten. He often got cranky, but he knew color science.

Dan

[Round Tuit]

The L* in the L*a*b* colour space is the lightness channel, so it will only change that value; black has a value of 0 and white a value of 100. A lot of professional retouchers convert their images to this colour space and work only with the L* channel for dodging and burning because it does not change hue or saturation.

Actually, this is wrong. Changing the L value changes the hue,saturation and brightness. The easiest way to demonstrate this is to open the colour picker, place the cursor in the L box of the Lab values display and set the L value to 100. Use the down arrow on your keyboard to step down the value from 100 to 1 and watch the convoluted path that the little circle on the colour patch takes to go from close to the upper left corner to the bottom right corner.

[Manfred M]

Manfred,

I don't think that's right.

I understand why people use Lab, and you'll probably recall that I posted images here a long time ago showing that tonality adjustments using L and the same adjustments using RGB mode and a luminosity blend mode produced essentially identical results.

The RGB values are brightness of those channels, not chroma. You can see this if you pull down midtones with a curve and switch to the color blend mode. The RGB values will remain nearly unchanged. Now, change the same curve to a normal blend mode or a luminosity blend mode. In both cases, the values of the RGB channels will drop because all three colors have been darkened.

The saturation value tabled by Photoshop, as discussed above, is not the same as perceived saturation, that is, perceived color intensity. It's calculated in various ways, but basically as a ratio of chroma to luminance, so darkening an image with L mode or with a luminosity blend will necessarily increase calculated saturation.

I don't know enough color science to understand the relationships among these quantities. However, in practical terms, it seems clear. if you want to change tonality without changing the perceived intensity of color, use Lab mode or RGB mode with a luminosity blend mode. Conversely, if you want to affect the perceived intensity of color without changing luminance, use RGB mode with a color blend mode.

We need the guy from Texas who used to post here, whose name I have forgotten. He often got cranky, but he knew color science.

Dan

Yes, Ted knew his colour theory and studied it (he's the only person that I knew that actually ordered copies of the various standards). That being said, I found that while he "knew" the theory, he had no idea on its application and effect in real world use. At times he would split hairs on something that effectively had little or no real world impact.

[Manfred M]

Actually, this is wrong. Changing the L value changes the hue,saturation and brightness. The easiest way to demonstrate this is to open the colour picker, place the cursor in the L box of the Lab values display and set the L value to 100. Use the down arrow on your keyboard to step down the value from 100 to 1 and watch the convoluted path that the little circle on the colour patch takes to go from close to the upper left corner to the bottom right corner.

I saw that with my testing as well, but what I don't know is how the colour models (L*a*b* versus HSB (and the closely related HSV) work. The L* (in L*a*b*) and B (in HSB) channels do very much the same thing.

When I work in the L*a*b* colour space and I adjust things in the L* channel, there are no changes in the a* or b* channels. Which is what I would expect. I see all three values change in HSB (which is not 100% intuitive).

What none of us know is the underlying math in the implementation of these Photoshop functions. Is the mapping from one colour space to the other working the way it should? This would not surprise me that it is not, as Photoshop's mapping from ProPhoto RGB and Adobe RGB to CMYK was problematic in the past and the colours looked "wrong" after conversion. I would have to converted from one of these wide-gamut colour spaces to sRGB to get acceptable looking CMYK output. I have not tested this lately, so I don't know if this has been fixed.

Because of these issues, I tend to work based on what the image ends up looking like, regardless of what the data says

[DanK]

I did a little more reading about color models tonight, and it's very complex. I decided it isn't worth my while to spend the hours it would take to learn enough to understand all of what is going on.

However, one thing seems clear to me: when we colloquially say that changing tonality in RGB normal mode changes saturation, while changing tonality in RGB luminosity mode or by changing L in Lab doesn't change saturation, we actually aren't talking about saturation in a formal sense. We are talking about colorfulness, or maybe chroma. I think the former.

For practical purposes, that's what we should be talking about, and the advice is simple: to avoid changing colorfulness, or whatever it is, either work in Lab mode or use a luminosity blending mode. It makes no difference which you choose.

BTW, Lab and L*a*b* aren't quite the same. I am guessing that what Adobe uses is L*a*b*, but it's often referenced simply as Lab. Just my guess.

[Manfred M]

I agree Dan. The subject is quite complex and I have been making my way through the three books that Dan Margulis has written on the subject of L*a*b*, over a number of years. It is complex and not easy reading. Colour and colour models are very complex subjects even before we add the vagrancies of the human visual system.

My understanding that L*a*b* is the correct name for the colour space and while we often see it written as Lab, that is technically incorrect. I have also seen saturation described as colourfulness.

Regardless, what is important is that there are many ways to describe a particular colour and there are subtle (mathematical) differences in how the specific colours can be described.

Ultimately, I suspect our aims are similar - get the image looking the way we want it to. No wonder that there are some people that stick to B&W...