3

Do the graticules in the curves editor have any absolute meaning?

The curves editor (in Adobe Camera Raw/Photoshop/Lightroom, probably all other image manipulation programs) has divisions along the X and Y axis. Do the values across those divisions have any absolute meaning?

For example some people claim that one square equals one stop. I strongly doubt that because notions like "a stop" only make sense when being scene referenced. The curve transformation would have to be applied in the input color space, but in general the photo editor is not even aware of the input color space. It's working in a device-independent working space.

Adobe Camera Raw might be an exception here because it has access to the device input profile, but I doubt it's doing anything different than Photoshop because the histogram displayed in the curved editor (and everywhere else) seems to have been de-linearized (TRC curve applied).

Also if Adobe Camera Raw would be working in device-dependent color space, and the gradations in the curve editor would have any scene-referened meaning, they would be different for every camera (because every camera has a different dynamic range), which is not the case.

If the gradations in the curve editor have no inherent meaning, how were they chosen?

  • Some modules in Darktable have the option of displaying a logarithmic scale. When pondering the curves, it is worth keeping in mind that RAW data is typically captured linearly and the Human Visual System (HVS) experiences light approximately logarithmiclly. So there will always be some friction between any encoding and some possible uses. – user50888 Dec 7 '17 at 17:54
1

The X and Y axis of the curve is the histogram range of [0..255]. See the gray scale tonal graduations shown there with it? X is input and Y is output, and the purpose of the curve line is to map an input RGB value to an output RGB value.

Photoshop shows four divisions, that I would call RGB 0, 64, 128, 191, 255. If you are currently showing image data, that image histograms [0..255] values are also superimposed there.

It absolutely is NOT stops. Stops are logarithmic, but the curve scale is linear RGB [0..255]. However, the histogram data itself is gamma encoded, data to the power of 1/2.2 (roughly square root).

So perhaps it does superficially seem to approximate stops, partially. Regarding stops, if linear data (which is Not shown), one stop down from 255 is 128, and 2 stops down is 64.

But in gamma 2.2 data (which is shown), one stop down from 255 is 186 (at 73%), and two stops down is 136 at 53%. We could call that close.

But three stops down is 99 at 39%, and four stops is 72 at 28%, neither very close to the graduations.

  • 1
    What color space is the data that the curves adjustment filter works on? Your answer implies it not the device-dependent input space (which empirically I agree with), but you still have not answered what is the working color space. There are no "RGB values", RGB values only have meaning in a particular color space. It doesn't seem to be the document working color space (though more testing needs to be done to ascertain that). You seem to be implying a pure gamma TRC of 2.2. Do you have any reference? Are you sure it's a pure gamma TRC, or is it the very close linear+power sRGB-like TRC? – Aram Hăvărneanu Dec 7 '17 at 18:06
  • Why complicate the obvious? It is of course the selected images color space. Maybe I should not have said RGB. The X and Y axis ARE the scale of [0..255] RGB, UNLESS you selected Graycale or CMYK or Lab, in which case it is their [0..255] scale, and shows their [0..255] image data. – WayneF Dec 7 '17 at 18:17
  • Why is it "of course" in the image color space? Do you have a reference for that? This does not follow from your answer, as you have specified a 2.2 gamma TRC. ProPhoto RGB, a very common working space has a gamma of 1.8, yet the curves adjustment seems to work independent of that with the same result. Also, most people use sRGB, with a TRC that is only approximatively equal to a pure gamma 2.2 one (sRGB is linear in the shadows and uses a linear+power combination). – Aram Hăvărneanu Dec 7 '17 at 18:29
  • Also, RGB, Graycale, CMYK, Lab, etc are color models, not color profiles. You still need a specific color profile for each color model to describe color. – Aram Hăvărneanu Dec 7 '17 at 18:30
  • It shows the image data and the histogram, whatever you have. I don't use gamma 1.8, but if that's what you have, the image data (arranged into a 0..255 histogram) is all there is to show. X Y shows the [0..255] values. If you have 16 bit data, the graphic histogram can still only show 256 pixels wide. – WayneF Dec 7 '17 at 19:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.