CMOS vs CCD and colour profiles a quick and dirty comparison

[eronald]

Hi -

 Could you explain what you mean by "matching" cameras?
 When you talk about DE, as you say you don't match lightness, is this effectively Delta AB?

Edmund

By the way, I did a quick comparison of IQ3100 vs A7r-II color response yesterday from Doug's raw files. I'd rather compare the IQx50 with the 645z as they're supposed to have the same sensor but I have no color checker raw shot of that at hand. What I could see though is that the "over-sensitive blue problem" with the A7r-II and Pentax 645z is not there with the IQ3100. This could be to a different CFA, or simply that the sensor glass cuts off some blue and not only IR. In any case a nicer hardware response to work with for color profile designers.

When it comes to matching cameras with profile I usually say that you can bring them very close, and indeed you can. However there will still be residual differences, and when you apply LUT relaxation to improve gradients the differences increase. For saturated colors there can easily be a 3-4 DE difference.

Another aspect is that many profile makers don't care to correct lightness with the LUT, and indeed per default DCamProf doesn't either. One reason for this is that colorcheckers are rarely evenly lit so lightness inputs are not to be trusted. Another reason is that lightness errors are the least disturbing, despite being easier to spot than hue/saturation in a quick A/B swap, and risking to break gradient performance just to correct lightness is often a bad idea. This means that lightness of saturated colors can differ quite much between cameras even when profiled for the same setup.

[ErikKaffehr]

Hi,

Delta E normally includes lightness: https://en.wikipedia.org/wiki/Color_difference

Best regards
Erik

Hi -

 Could you explain what you mean by "matching" cameras?
 When you talk about DE, as you say you don't match lightness, is this effectively Delta AB?

Edmund

[eronald]

Hi,

Delta E normally includes lightness: https://en.wikipedia.org/wiki/Color_difference

Best regards
Erik

Are you using delta E(76) or delta E(94)?

Edmund

[ErikKaffehr]

Hi Edmund,

I am not using Delta E these days, as I don't try to make numerical comparisons. When doing comparisons I mostly used CIEDE 2000.

The L value makes comparisons a bit messy, as tonal reproduction curve is intended to be slightly unlinear. I mostly used "Patch Tool" for comparisons.

Are you using delta E(76) or delta E(94)?

Edmund

[eronald]

Hi Edmund,

I am not using Delta E these days, as I don't try to make numerical comparisons. When doing comparisons I mostly used CIEDE 2000.

The L value makes comparisons a bit messy, as tonal reproduction curve is intended to be slightly unlinear. I mostly used "Patch Tool" for comparisons.

Erik,

 I am sorry, the quote was from Torger, and the question *for him* relative to *his* quote. I would like to know what DE measure *he* is using when he talks about 3-4 DE.

Edmund

[torger]

I'm using CIE DE2000, and when analyzing I often split it up to look at hue, lightness and chroma separately. It's not an exact measure though, especially in the more extreme ranges. When I was referring to a 3-4 DE difference it's just to indicate the scale of differences so which standard that is used doesn't matter that much.

The numbers included lightness, but I don't think "not correcting lightness + LUT relax" compared to "correcting lightness + LUT relax" does that much of a difference since the relax brings it closer to a matrix solution anyway, it depends on how strong relaxation is. All this can be controlled with DCamProf, and the default relax is quite strong.

The goals of a general-purpose profile and a reproduction profile is quite different. I aim for general-purpose profiles and then it doesn't make much sense to try to get colors super-correct for a particular target and lighting condition and introduce strong LUT stretches to do so. Instead the LUT is kept quite relaxed to guarantee good gradients in any condition, and then "the magic" takes place in applying the "film curve" and subtle subjective adjustments in the "look" (this is also applied via a LUT).

I'd love to put some more effort into the design bits and make a GUI etc, and I have still much to learn about this, but it's a huge project and you really can't sell the stuff in any numbers, so it's not strange that there's not much profile design software around... all stuff the manufacturers do is in-house with their own custom software, and they don't talk about how they do things.

Could you explain what you mean by "matching" cameras?
 When you talk about DE, as you say you don't match lightness, is this effectively Delta AB?

[eronald]

I'm using CIE DE2000, and when analyzing I often split it up to look at hue, lightness and chroma separately. It's not an exact measure though, especially in the more extreme ranges. When I was referring to a 3-4 DE difference it's just to indicate the scale of differences so which standard that is used doesn't matter that much.

The numbers included lightness, but I don't think "not correcting lightness + LUT relax" compared to "correcting lightness + LUT relax" does that much of a difference since the relax brings it closer to a matrix solution anyway, it depends on how strong relaxation is. All this can be controlled with DCamProf, and the default relax is quite strong.

The goals of a general-purpose profile and a reproduction profile is quite different. I aim for general-purpose profiles and then it doesn't make much sense to try to get colors super-correct for a particular target and lighting condition and introduce strong LUT stretches to do so. Instead the LUT is kept quite relaxed to guarantee good gradients in any condition, and then "the magic" takes place in applying the "film curve" and subtle subjective adjustments in the "look" (this is also applied via a LUT).

I'd love to put some more effort into the design bits and make a GUI etc, and I have still much to learn about this, but it's a huge project and you really can't sell the stuff in any numbers, so it's not strange that there's not much profile design software around...

Ah. Thx for your reply.

My opinion on GUI is that ergonomics is a different skill, and a good way for a core author  to do things is to make a set of batch-callable small routines and let people wrap a GUI around them. With a web backend calling things is easy, and the frontend is then web programming. I once did a demo for Barbieri by doing an Argyll wrapper with shell scripts and a web frontend, yielding a pushbutton profiler. And Iliah Borg also wrote something using Livecode as a wrapper...

One reason for the absence of *commercial* software is that there are a bunch of patents, especially by Kodak and people are very afraid of them.

Edmund

[AlterEgo]

I'd love to put some more effort into the design bits and make a GUI etc

better to spend possible effort on core functionality and documentation...

[eronald]

better to spend possible effort on core functionality and documentation...

Yes, I agree. Argyll and LittleCMS are good example of programs adopted by the color community because they are scriptable and documented.

BTW the real market is in cine grading and matching cameras. Cine guys have a budget for every project, still guys have a single software budget for their whole hardware kit.

Edmund

[ErikKaffehr]

Hi,

I shot some real world examples. These images have same processing, in Lightroom, using the profiles generated by Anders Torger's DCamProf.

P45+ on the left A7rII on the right.

Best regards
Erik

[ErikKaffehr]

Hi,

Here is another one. P45+ on the left A7rII on the right.

Best regards
Erik

[Bart_van_der_Wolf]

Hi,

I shot some real world examples. These images have same processing, in Lightroom, using the profiles generated by Anders Torger's DCamProf.

Hi Erik,

What's left and what's right, IOW what are we comparing with?

Left is more (on sRGB displays too much?) saturated, and right is more controlled and a bit less yellow.

Of course that is a basis for further processing, so the differences could be made smaller by not using the same settings, but for a comparison one attempts to change as few variables as possible, in this case only the profile.

Cheers,
Bart

[eronald]

Hi,

I shot some real world examples. These images have same processing, in Lightroom, using the profiles generated by Anders Torger's DCamProf.

When one looks at the full-scene images, the one on the left is alive, the one on the right is a zombie.

Edmund

[ErikKaffehr]

Hi Bart,

Left is P45+ and right is A7rII. I was in a great hurry posting those images, so I forgot to write which is which. Sorry!

Here the intent was to use a real world subject and keep everything the same, except profiles.

Tim Parkin has some interesting discussion on the issue: https://www.onlandscape.co.uk/2012/02/the-myth-of-universal-colour/

But that web site is not available right now.

I will check out some other profiles later and perhaps also RawTherapee.

The raw images will be posted, probably later today, so if anyone wants to play C1 or something else…


Best regards
Erik

Hi Erik,

What's left and what's right, IOW what are we comparing with?

Left is more (on sRGB displays too much?) saturated, and right is more controlled and a bit less yellow.

Of course that is a basis for further processing, so the differences could be made smaller by not using the same settings, but for a comparison one attempts to change as few variables as possible, in this case only the profile.

Cheers,
Bart

[ErikKaffehr]

Hi Edmund,

P45+ is the one alive and A7rII is the zombie.

Best regards
Erik

When one looks at the full-scene images, the one on the left is alive, the one on the right is a zombie.

Edmund

[eronald]

I used to show people two prints of a raw steak.

One print made with standard Epson profiles, the other with my custom profile.

One image looked like something you would buy, the other like something you would throw away.

Edmund

Hi Edmund,

P45+ is the one alive and A7rII is the zombie.

Best regards
Erik

[Bart_van_der_Wolf]

When one looks at the full-scene images, the one on the left is alive, the one on the right is a zombie.

Hi Edmund,

That's what happens, if different profiles for different lenses+cameras/sensors are used with the same post-processing. The real question (and not the main goal of the thread) is of course which image can be made to look better, after using the data we get out of the Raw conversion/demosaicing stage.

I prefer the more saturated look (but not the oversaturation of some tones, nor the overall yellow tone), but that is easy to also achieve with the image on the right because it seems to have a well managed tonality to begin with. The question becomes whether the image on the left still has data to allow some post-processing of saturated detail or is it all bunched up in similar values beyond the ability to create subtle tone differences.

Also, don't forget that one of the benefits of Medium Format is a potentially higher MTF, because image detail is projected larger on the sensor, and needs less magnification for output to a similar size.

Cheers,
Bart

[eronald]

Bart,

 I'm sure you're right. As for me - I simply don't use Lightroom
 Not using Lightroom is the single quickest way to make one's pictures better - if one can locate them

Edmund

Hi Edmund,

That's what happens, if different profiles for different lenses+cameras/sensors are used with the same post-processing. The real question (and not the main goal of the thread) is of course which image can be made to look better, after using the data we get out of the Raw conversion/demosaicing stage.

I prefer the more saturated look (but not the oversaturation of some tones, nor the overall yellow tone), but that is easy to also achieve with the image on the right because it seems to have a well managed tonality to begin with. The question becomes whether the image on the left still has data to allow some post-processing of saturated detail or is it all bunched up in similar values beyond the ability to create subtle tone differences.

Also, don't forget that one of the benefits of Medium Format is a potentially higher MTF, because image detail is projected larger on the sensor, and needs less magnification for output to a similar size.

Cheers,
Bart

[Jack Hogan]

When one looks at the full-scene images, the one on the left is alive, the one on the right is a zombie.

Agreed.  The problem is that the difference appears to be almost entirely in saturation, which appears to have become an arbitrary parameter in current profiles since DNG 1.4.  Hey, the minute you buy into the idea that all a profile needs to do linearly is plop you into perceptual intent projection lines, the sky is the limit and Ken Rockwell is perceptually colorimetric.

Jack

[Jack Hogan]

Not using Lightroom is the single quickest way to make one's pictures better - if one can locate them

I don't use Lightroom precisely because I can never locate them in my three working computers.

Jack