How do scanners color correct C-41 negatives?

[PhilBurton]

I appreciate all your efforts.
Correct me if I am wrong, but any attempt to deal with a particular film's mask is unnecessary and not well founded, because it is the combination of the mask (which varies with the image) and the image itself that determines the matter.

Is my understanding too naive? Whatever the image, the settings in any software to remove the mask should be the same. Those settings may vary according to the specific Kodak (or other brand) film. Hopefully the settings don't vary from roll to roll of a specific film, meaning that processing variations from roll to roll aren't significant. Does the age of the film have an impact on the color mask?

How do commercial printing machines deal with variations in print mask?

 

[MattKing]

Film + mask = a single standard.
The standard is the one defined by the colour paper.
The software doesn't "remove" the mask. The software both inverts the image and converts it from the "colour space" that the paper expects to see to one you want to see.
Any film specific settings should be related to the particular characteristics (contrast, saturation) of the film.
Optical colour printers never had to adjust from film to film, because they were all designed to be exposed to the same colour paper.

 

[Derek L]

Let me try a more empirical approach. Suppose I want to emulate the look of, for example, Fuji Frontier scanners, which I understand were designed to reproduce the look of a RA-4 print, more or less. (Can anyone confirm this?)

The obvious way to do this is to shoot a big color chart (cheap references with hundreds of patches exist), send the film in to be developed and scanned, then create a 3d transform (3D LUT) that maps the negative values from the DSLR scan to the scanner's colors. This should give great results for DSLR scanning other negatives from the same stock.

However, the problem is that I think the scanner may be doing automatic color correction for each negative. So the density-to-color map it implements may change negative to negative, and in particular it may do something radically different to a shot of a color chart against a black background than it would to a typical scene.

Could someone who knows how these scanners work please comment? If this method is tractable, I will do this and post the results and code.

Edit: I guess the right thing to ask for is a "straight scan" with no color correction done in-scanner. Does anyone know if these produce decent quality? It would be sort of ridiculous to go through all this trouble for a mediocre end product.

 

[MattKing]

Let me try a more empirical approach. Suppose I want to emulate the look of, for example, Fuji Frontier scanners, which I understand were designed to reproduce the look of a RA-4 print, more or less. (Can anyone confirm this?)

No...
They were designed to either:
1) produce a digital file that, when printed with a digital printer designed to print digitally to RA-4 paper, would produce a print on RA-4 paper that was similar to an optical printer used to print on RA-4 paper; or
2) produce a digital file that, when printed with a digital printer designed to print digitally to inkjet paper, would produce a print on injet paper that was similar to an optical printer used to print on RA-4 paper.
The two options are quite different.

 

[Derek L]

No...
They were designed to either:
1) produce a digital file that, when printed with a digital printer designed to print digitally to RA-4 paper, would produce a print on RA-4 paper that was similar to an optical printer used to print on RA-4 paper; or
2) produce a digital file that, when printed with a digital printer designed to print digitally to inkjet paper, would produce a print on injet paper that was similar to an optical printer used to print on RA-4 paper.
The two options are quite different.

I don't understand. In a properly color managed workflow, shouldn't these be the same? That is (up to gamut issues, etc.) whatever I see on my monitor should be essentially what is printed by either the inkjet or RA-4 digital process.

After all, isn't "what you see is what you get" the whole point of color management?

 

[PhilBurton]

No...
They were designed to either:
1) produce a digital file that, when printed with a digital printer designed to print digitally to RA-4 paper, would produce a print on RA-4 paper that was similar to an optical printer used to print on RA-4 paper; or
2) produce a digital file that, when printed with a digital printer designed to print digitally to inkjet paper, would produce a print on injet paper that was similar to an optical printer used to print on RA-4 paper.
The two options are quite different.

I will admit freely that I have never done color printing myself, so I'm completely ignorant of actual color printing with actual negative films. However ......

If I am scanning color negative film, why would I not want a "neutral" positive image as a result, as though I had created that digital image with a (Nikon|Canon|Sony) DSLR? Yes, i realize that there is the grain inherent to film which is not present in a digital file. But aside from that, would I not want a completely color-corrected digital image? er

Is it written somewhere that a digital scan of a color negative must be printed to look as through the print came from a film printer that projected the negative onto paper?

If I scan a slide, isn't it an objective to set the scanner to correct for the characteristics of specific films, such as Kodachrome, such that i can't look at a digital image and say, "Oh, that was obviously scanned from Kodachrome 25," or "Ektachrome X?"

Again, I'm probably showing my complete ignorance here. So what am I missing that is obvious to everyone else?

Phil

 

[MattKing]

Those Fuji Frontier scanners were designed for the specific printing machines they were attached to.
The files they produce(d) were/are also distributed to customers, but they don't have a "look" - they are designed to be printed on Fuji printers.

 

[MattKing]

I will admit freely that I have never done color printing myself, so I'm completely ignorant of actual color printing with actual negative films. However ......

If I am scanning color negative film, why would I not want a "neutral" positive image as a result, as though I had created that digital image with a (Nikon|Canon|Sony) DSLR? Yes, i realize that there is the grain inherent to film which is not present in a digital file. But aside from that, would I not want a completely color-corrected digital image? er

Is it written somewhere that a digital scan of a color negative must be printed to look as through the print came from a film printer that projected the negative onto paper?

If I scan a slide, isn't it an objective to set the scanner to correct for the characteristics of specific films, such as Kodachrome, such that i can't look at a digital image and say, "Oh, that was obviously scanned from Kodachrome 25," or "Ektachrome X?"

Again, I'm probably showing my complete ignorance here. So what am I missing that is obvious to everyone else?

Phil

You would want a file that is usable on your equipment.
The Fuji (or Kodak, or Pakon, or ....) scanners are designed to output files used in labs.
Those labs have printers and those printers have characteristics. Scanners from Fuji are likely to be at least optimized for creating files for printing on Fuji printers.
Most importantly though, it makes no sense to set a lab scanner up to make all films look the same, because it is probably impossible to differentiate between characteristics of the film and characteristics of the subject - otherwise all sunset pictures would come out looking like mid-day.
If you want to customize your own, individual scanning preferences - go for it. As you are essentially the only customer for your scans, your customization makes sense.
Films are designed for optical printing on RA-4 paper. The standardization has been done. You may very well want to convert the results for different sorts of presentation - your screen, website posting, digital RA-4 printing, digital inkjet printing, etc. - but the way to do that is through colour spaces, not changing the scanning process.

 

[Adrian Bacon]

You would want a file that is usable on your equipment.
The Fuji (or Kodak, or Pakon, or ....) scanners are designed to output files used in labs.
Those labs have printers and those printers have characteristics. Scanners from Fuji are likely to be at least optimized for creating files for printing on Fuji printers.
Most importantly though, it makes no sense to set a lab scanner up to make all films look the same, because it is probably impossible to differentiate between characteristics of the film and characteristics of the subject - otherwise all sunset pictures would come out looking like mid-day.
If you want to customize your own, individual scanning preferences - go for it. As you are essentially the only customer for your scans, your customization makes sense.
Films are designed for optical printing on RA-4 paper. The standardization has been done. You may very well want to convert the results for different sorts of presentation - your screen, website posting, digital RA-4 printing, digital inkjet printing, etc. - but the way to do that is through colour spaces, not changing the scanning process.

+1

CIE XYZ is typically the connection colorspace once things are digitized, so once you’ve got it digitized and worked out how to cleanly get to XYZ, you can convert to any other color space you need to. In a color managed environment, this is all done automatically for you on the fly, so that you can even see anything on your display. I imagine that the frontiers and noritsus of the world largely work this way under the covers. If they didn’t, then they’d have to have custom displays, etc, as pretty much everything manufactured in the last 20+ years that deals with color is designed to work within ICC standards.

How to cleanly get to XYZ? That is specific to the hardware you’re using to get your digitized samples. Once you have that worked out, it’s pretty much the same process for all films. Individual emulsions might need tweaks to contrast, hue, or saturation, but the overall process is the same.

 

[PhilBurton]

You would want a file that is usable on your equipment.
The Fuji (or Kodak, or Pakon, or ....) scanners are designed to output files used in labs.
Those labs have printers and those printers have characteristics. Scanners from Fuji are likely to be at least optimized for creating files for printing on Fuji printers.
Most importantly though, it makes no sense to set a lab scanner up to make all films look the same, because it is probably impossible to differentiate between characteristics of the film and characteristics of the subject - otherwise all sunset pictures would come out looking like mid-day.
If you want to customize your own, individual scanning preferences - go for it. As you are essentially the only customer for your scans, your customization makes sense.
Films are designed for optical printing on RA-4 paper. The standardization has been done. You may very well want to convert the results for different sorts of presentation - your screen, website posting, digital RA-4 printing, digital inkjet printing, etc. - but the way to do that is through colour spaces, not changing the scanning process.

Matt,

You may be giving me too much credit here. Maybe I'm being simple-minded, but all I want is to be able to scan such that the resulting digital positive file has true, neutral colors, preferably in the aRGB color space. So I would want my scanning software to either have pre-sets for different C-41 (and I also have some old C-22) films, or else have some sort of guide for how to scan just the film base and arrive at the right settings. This is the same that I would want if scanning slides.

 

[Derek L]

Matt,

You may be giving me too much credit here. Maybe I'm being simple-minded, but all I want is to be able to scan such that the resulting digital positive file has true, neutral colors, preferably in the aRGB color space. So I would want my scanning software to either have pre-sets for different C-41 (and I also have some old C-22) films, or else have some sort of guide for how to scan just the film base and arrive at the right settings. This is the same that I would want if scanning slides.

I can almost guarantee you that you do not want the true, scene-referred colors. These are judged by most people to look "flat" compared to what the eye experiences in real life and generally displeasing. There's a reason why raw file converters for digital cameras apply custom color profiles to the raw files, even though they are supposed to represent "digital negatives."

But don't take my word for it. You can easily profile a digital camera to give scene-referred color (up to small amounts of error inherent in the measurement process).

 

[MattKing]

So I would want my scanning software to either have pre-sets for different C-41 (and I also have some old C-22) films, or else have some sort of guide for how to scan just the film base and arrive at the right settings. This is the same that I would want if scanning slides.

You should not be looking for different scanning pre-sets for different C-41 films - they are all designed to print to exactly the same target - RA-4 paper. Which means they are designed to all scan in the same way.
The different films might have different characters - one may emphasize saturation, another may emphasize contrast, a third might have stronger red response, a fourth might favour greens, etc., etc. - but the differences between the films are subjective differences that guide people when they are choosing a film in the first place.
If your goal is to convert images from all sorts of different films into one sort of image, the place to do that is in post-processing - not scanning.

 

[Ted Baker]

The obvious way to do this is to shoot a big color chart (cheap references with hundreds of patches exist), send the film in to be developed and scanned, then create a 3d transform (3D LUT) that maps the negative values from the DSLR scan to the scanner's colors. This should give great results for DSLR scanning other negatives from the same stock.

This will result in another fudge. The more successful you are in matching the colours in the reference chart, the more reason you should have used a DSLR in the first place.

Why?

You do not want a complete system that gives colourmetric correct results from the scene to your final output! Read the Giorgianni and Madden for examples. There would be no point in using different films other than ISO ratings...

Also you need to remember that different exposure changes the results on the camera negative, there is a lot of clever chemical engineering built in the negative that controls saturation for example. Different exposure indices give different results. I could go on...

Could someone who knows how these scanners work please comment? If this method is tractable, I will do this and post the results and code.

Kodak Cineon is the best to study, It is unchanged for 20 years works perfectly but is complex to understand, the equipment has 7 figure budgets etc. I think you will find fuji frontier and Noritsu use the same methodology but they don't disclose any of it so that is just a guess. Cineon is just each step of the analogue system mapped to the digital domain. Also you will see how they map modern digital photography, and stuff that is digitally generated by computer for special effects work into this pipeline.

In summary it is print centric, that you should be able to use the same print stock (i.e. RA4 paper or Vison 3 print stock) with digital printer, or traditional optical print and get similar results.

There is also PhotoCD which is now defunct but find some documentation worth reading.
Then there are lots of other fudges some better than others...

 

[Alan Edward Klein]

You should not be looking for different scanning pre-sets for different C-41 films - they are all designed to print to exactly the same target - RA-4 paper. Which means they are designed to all scan in the same way.
The different films might have different characters - one may emphasize saturation, another may emphasize contrast, a third might have stronger red response, a fourth might favour greens, etc., etc. - but the differences between the films are subjective differences that guide people when they are choosing a film in the first place.
If your goal is to convert images from all sorts of different films into one sort of image, the place to do that is in post-processing - not scanning.

So, you should be able to use Epson scan software flat with EPson scanners?

 

[Ted Baker]

After all, isn't "what you see is what you get" the whole point of color management?

Color management only works if you have a linear model, say for example you have a colour that is

25% Red, 50% Green and 75% Blue, and you want to use a linear matrix to change the colours because you are using new primaries. The maths works fine if 75% is 3 times more intense that 25% it does not work if you don't know what the relationship between 75% and 25% is because they are just some arbitrary values without real meaning.

 

[Derek L]

This will result in another fudge. The more successful you are in matching the colours in the reference chart, the more reason you should have used a DSLR in the first place.

Why?

You do not want a complete system that gives colourmetric correct results from the scene to your final output! Read the Giorgianni and Madden for examples. There would be no point in using different films other than ISO ratings...

I think I was unclear. I certainly do not want to reproduce the scene-referred color. I propose using the DSLR as a budget densitometer and reconstructing the (e.g. Noritsu) scanner's density to color map.

Color management only works if you have a linear model, say for example you have a colour that is

25% Red, 50% Green and 75% Blue, and you want to use a linear matrix to change the colours because you are using new primaries. The maths works fine if 75% is 3 times more intense that 25% it does not work if you don't know what the relationship between 75% and 25% is because they are just some arbitrary values without real meaning.

I would appreciate a reference for this claim, because as far as I know it is absolutely not true that color management only works with linearly encoded color. At the very least, ICC profiles can easily handle gamma adjustments, which are nonlinear.

 

[Photo Engineer]

So, you should be able to use Epson scan software flat with EPson scanners?

Yes, I do!

 

[Ted Baker]

I think I was unclear. I certainly do not want to reproduce the scene-referred color. I propose using the DSLR as a budget densitometer and reconstructing the (e.g. Noritsu) scanner's density to color map.

OK that work either (or at least it is still another fudge) as a one step process or mapping because it not a 1 step process or mapping... It it a multistage process or mapping. If you try and work out all the permutations you will need it is huge.

I would appreciate a reference for this claim, because as far as I know it is absolutely not true that color management only works with linearly encoded color. At the very least, ICC profiles can easily handle gamma adjustments, which are nonlinear.

Its implied. Of course there is provision for gamma encoding, that allows you add that INTO your calculations. That is the whole point! A gamma curve or any tone curve is from a reference line which is straight linear line.

 

[Derek L]

OK that work either as a one step process or mapping because it not a 1 step process or mapping... It it a multistage process or mapping. If you try and work out all the permutations you will need its huge.

An IT8 target has 288 patches. With the correct interpolation function it should get good results. People profile cameras/scanners with it and do fine.

Also, a gamma correction is a power law, not linear, so perhaps we are talking past each other.

 

[Ted Baker]

Also, a gamma correction is a power law, not linear, so perhaps we are talking past each other.

ICC profiles have several ways to describe a non linear curve, but it is implied that the description is using a straight line where 10 is 10 times bigger than 1 as a reference.

 

[Derek L]

OK. So I guess your claim is really that ICC profiles are all 3x3 matrix profiles?

But this is not the case. I can make lookup table ICC profiles in the open source ArgyllCMS software, for example. So nonlinearly of the transform (before or after the gamma adjustment, whatever) won't matter for color management.

 

[Ted Baker]

OK. So I guess your claim is really that ICC profiles are all 3x3 matrix profiles?

No such claim. All I am really saying is for any measurement to make sense you have to have some reference. That implied reference line is a linear scale.

An IT8 target has 288 patches. With the correct interpolation function it should get good results. People profile cameras/scanners with it and do fine.

Its still a fudge, I am not saying it may be sufficient, and helpful, but IT WILL NOT GET YOU ALL THE WAY THERE!

 

[Derek L]

I guess I don't understand what you mean by "color management only works if you have a linear model," then.

I agree that if you don't know the profile being used, then you can't do anything. But presumably the Fuji engineers know how the values in their custom, sRGB-like space can be converted to linear light measurements. If that is the case, there should be no problems with color managing their workflow in principle. (Whether they implemented such a feature is of course another question.)

Also, yes, of course the 288 patch profile would be necessarily imperfect. But I can test model fit after the fact, and data online indicates that even after ~50 well-chosen patches, the differences are barely perceptible (as modeled by delta-E).

 

[Adrian Bacon]

I can almost guarantee you that you do not want the true, scene-referred colors. These are judged by most people to look "flat" compared to what the eye experiences in real life and generally displeasing. There's a reason why raw file converters for digital cameras apply custom color profiles to the raw files, even though they are supposed to represent "digital negatives."

But don't take my word for it. You can easily profile a digital camera to give scene-referred color (up to small amounts of error inherent in the measurement process).

The final output should not generally be scene referred, but you do want that for the input because all of the color algorithms operate best with gamma 1.0 data (scene referred is gamma 1.0, with a correctly exposed 18% grey card being 0.18, or 11796 if using integers). Most color spaces gamma encode the actual samples, usually 1.8, 2.2, or 2.4, depending on the colorspace. The color environment always removes the gamma so it’s 1.0 before doing any operations, then converts back to the gamma of the color space. The reason why I picked ACES for my color space is because it’s already gamma 1.0, less conversions, less rounding errors due to being constantly converted every time you modify it.

In short, you do want linear scene referred for your input, but it’s totally acceptable and expected that you’ll deviate from that for your output.

 

[Derek L]

The final output should not generally be scene referred, but you do want that for the input because all of the color algorithms operate best with gamma 1.0 data (scene referred is gamma 1.0, with a correctly exposed 18% grey card being 0.18, or 11796 if using integers).

One thing I don't understand is, if you've developed profiles for films to apply to scene-referred data, why not just skip the analog capture and use a profiled DSLR to get a scene-referred capture, then convert that? It's faster and cheaper.

I ask this half in jest—I understand there are properties other than color about film you might want to preserve. But you could, for instance, use any stock and make it look like any other. At this point you've erased all the color science that was engineered into the film and made it just another capture device. Don't get me wrong, that's technically impressive and I can't argue with the results, which I really like. It just strikes me as an odd way to go about scanning negatives.