4

I am getting the following very unpleasant visual effect from a party picture I am trying to post-process in RawTherapee. Basically, I would like to have the regions where the lights pointed to look white and white-ish (as they actually were in real life) — and not light-purple and dark-blue.

enter image description here

The picture was shot with my Nikon D750, and the camera (somehow) managed to fix this in the JPG it exported alongside the raw file.

I have tried to debug this, and it seems that for those specific pixels the red and blue channels are saturated (or almost saturated) whereas the green channel is completely unsaturated — any ideas why?

  • I thought I might be able to deal with this by adjusting the tint accordingly (for purple to green). That worked as expected for the highlights, but ruined the rest of my picture.

  • I believe the issue could be solved by having the red and blue channels "bleed" into the green channel when they approach saturation. But I found no way to do this in RawTherapee. Do you know how/whether this is possible?

  • Maybe I am having his issue because RawTherapee has no colour profile for my camera yet (I am thinking maybe it is normal for the sensor to just cut out the green for very strong colours). Any ideas how I could best get a colour profile?

  • Looks like a bug to me. – Mark Ransom Oct 22 '15 at 22:06
  • The comment you mad on saturation is what you see in the above image. What did the RGB curves look like before processing - eg out of camera jpg if avaiable or jpg processed in as "normal" a way if RAW only available? – Russell McMahon Oct 23 '15 at 6:05
  • @MarkRansom, lika a bug with what, RT or my camera sensor? – TheChymera Oct 23 '15 at 11:43
  • 1
    Adjusting the tint should be done locally, not on the whole image. I'm not sure if RT can do that, I use LR only. But hey, those blue and purple make the photo more party-ish! Call it "art" and you are done. And ask in Raw Therapee forums to ensure it's not a bug, your experience may be of help. – FarO Oct 23 '15 at 12:01
  • 1
    Can you show us the result of in-camera conversion? This would be helpful. – Euri Pinhollow Nov 15 '16 at 8:31
1

The reason for them to look so terrible is aggressive profiling which needs to be done to bring the image data recorded by camera closer to the reality (the bad form of colour sensing to be specific), and it's imperfection is exceptionally obvious with saturated objects. This is very apparent in photos of recent Canon DSLRs because they tend to require strong profiling and will show up with any digital camera if light source is saturated enough. If it wasn't the case the light would look good as long as it is inside colour space of output file and inside colour space of output device.

Most output devices cannot reproduce the colour which is produced upon simple colour conversion and it needs additional manipulation to look fine.

Here are two verions from same file, one from Adobe Camera RAW and one from RawTherapee with custom weaker (less saturated) colour profile.

There is a way of making saturated bright sources look good - give tonal curve a wide shoulder and it will look natural. (your tonal curve will be different from the one displayed, experiment to find the one which you need)

You may also try different highlight reconstruction settings and also reduce saturation globally but the main way of improving saturated highlight is still tonal curve (or highlight compression which acts similarly).

Related.

Also, recent versions of RT sure have a profile for D750 and it is contained in camconst.json . You may either update RT or copy just one file from installation folder.

  • If that's true, then it's a fundamental design flaw in the software. A proper image processing pipeline should perform computation using large enough variables to ensure that no truncation of intermediate results occurs. And if the final color falls outside the current color space, assuming you picked a reasonable rendering intent, you should get a color that's the closest match for the computed color that falls within the color space. If that doesn't happen, it's a bug. You should never suddenly see bright blue where the result should be grey. – dgatwood Nov 12 '16 at 21:35
  • @dgatwood: that's a fundamental flaw of recording technology, not the software. The closest match for a super saturated blue (which results from strong profiling required by cameras) is the most saturated blue of the LCD, there is nothing the software can do about it implicitly, every solution has it's effect on all images - even those which are fine. Think about it as aliasing: it is an algorithmically undetectable defect fixing which will affect even images which are fine. – Euri Pinhollow Nov 13 '16 at 8:26
  • @dgatwood: universally reducing the saturation of everything does not help because there are other photos which are fine and will be made more wrong with this change. – Euri Pinhollow Nov 13 '16 at 8:31
  • The point I was trying to make was that unless you're trying to do aggressive white recovery for some reason (which doesn't seem likely for a mostly-dark picture like the one posted above), the normal color-space conversions shouldn't result in artifacts, because white should still be roughly white in the final output space, and if it gets clipped on output, it should still clip to the nearest match, which is white. So if it ends up being a different color, it almost certainly means that there's a loss of precision in the computations that results in truncation. – dgatwood Nov 14 '16 at 18:29
  • @dgatwood: it is very unlikely for RT to have such a bug which makes whites saturated, it has very simple input processing. The only reason the light are saturated is because they are made this way with strong profile which camera requires. The camera does internally something similar to what I propose: it uses tonal curve with wide shoulder and also probably increases brightness quite a bit. – Euri Pinhollow Nov 15 '16 at 16:22
1

The fix is called Saturation&Value Blending option on the Tone curves.

At leasat for me it is the easiest compromise. NEFs from D5100.

Standard

Saturation/Value blending

  • Can you please extend your answer and give some examples – Romeo Ninov May 14 '17 at 18:59

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.