Nikonscan 4.0.3 C41 inversions and RA4 paper colour signature question

[albireo]

This question straddles the two realms of fully analogue and hybrid film photography and I'm not sure where to post it - so please mods move as appropriate.

I have been experimenting a lot with medium format colour negative film as of late. I am mostly a black and white photographer and I exclusively scan my negatives, and I had never been really happy with the colours I got from my scanned C41 negatives until recently.

I had experimented with a number of methods:

• Vuescan
• Vuescan 'advanced workflow'
• Vuescan Raw + Colourperfect plugin
• Vuescan Raw + NLP
• Vuescan Raw + Filmomat

All of the above resulted in -subjectively- only ok-ish (at best) or extremely problematic (at worst) colour inversions. I know some people would suggest manual inversions and curve tweaking in Photoshop as an added option, but I've never gotten results I really liked this way (I have a mild form of red-green colour blindness and don't fully trust my judgement when performing tiny tweaks to a 6-way colour slider) and I do not like spending more than 2 minutes per picture per image in Photoshop post-scanning,

So I was really after a method that does most of the work, leaving me with minor final adjustments (resize, crop, set black point, save to jpg). I couldn't find it. I was about to slowly move away from C41 and shoot more black and white and slide film when I decided, for fun, to install a really old, unsupported bit of software for my main dedicated film scanner (a Coolscan 8000ED): Nikonscan 4.0.3. It was very easy to install on my Windows 10 64bit machine and it runs surprisingly well. I decided to delve a bit deeper and rescan some of my C41 negatives.

I was astounded by what I was getting. Rich, beautiful colours that look 'right' to me. The results from even a completely automated C41 scan session with Nikonscan are - to my eyes - leaps and bound better than what I was getting before. It's hard to describe, and I will be posting some side by side samples when I have the time. For 90% of the negatives I've thrown at it (Gold, Fuji 400H, Ektar), I have been happy with the automated scan and done close to nothing in post-processing. Bliss.

As a side note - for those of you who attended the William Eggleston exhibition in Berlin, Germany, a few months ago, you will know exactly what kind of colour I'm talking about: not tacky, not oversaturated, not desaturated, but just 'right' - alive. The good news is, for those of you scanning, you can get those colours from simple, widely available C41 colour film with a proprietary Nikonscan inversion.

So after this preamble here's my question. I have been trying to gather more info about exactly what is the Nikonscan algorithm doing under the hood. I visited several film scanning communities asking for opinions on what I saw, and finding many other Coolscan users agreed with me and my observations. Many mentioned the Nikonscan code is sadly closed source and there should be a concerted effort to reverse engineer it and make it open source to save it. I agree it would be a pity to lose the capability to obtain these phenomenal results - again fully automated results - with close to no intervention on the user's side.

One comment, by a couple of users on the facebook Coolscan community - really caught my attention: these people mentioned that the tool was engineered to produce a result that closely matches the colour response of C41 film onto RA4 paper - as most film for the past decades had been designed to perform optimally on RA4 paper.

Do people have more info on this? Does the Nikonscan output looks good because it mimics RA4 paper response? Is it possible to design an algorithm to match or approximate RA4 paper? If the answer is yes, what are the characteristics of RA4 papers that are crucial in this context and that e.g. NLP and Filmomat are not accounting for?

 

[koraks]

I'm highly skeptical about what 'people' say on forums about scanning software mimicking RA4. The reason for this is that I've never come across any hints that these people had any inside information on the color algorithms and for the most part seemed to make an educated guess, only. Sometimes, that is good enough, but especially in this case, I don't think it helps any.

If this is a route you'd like to pursue, this blog may be useful to you: https://tinker.koraks.nl/photograph...look-at-kodak-and-fuji-digital-ra4-crossover/ Ignore the main issue addressed here, which is the analog vs. digital aspect of color papers. What's of interest to you, I imagine, are the curve shapes. You could try to recreate these and see if that gets you closer to what you want.

However...I consider this a red herring and I'd suggest just twisting some curves until you're happy with what you get. Ignore the whole RA4 paper thing. If you're scanning film, you're working with an abstraction of the actual negative, so you're never actually getting what the paper 'sees', and trying to approximate that will turn out to be a philosophical black hole and you'll run aground in metaphysical musings about the nature of reality and digital abstractions thereof. It's a dead-end street.

What does help, is to do a couple of conversions with the Nikon software and compare the output with raw scans made with your otherwise favorite software. Then replicate the behavior of the Nikon software, in which it'll be helpful to deliberately reflect on what you see. Maybe go through the tedious motions of shooting some color targets or test scenes to isolate specific hues and see how they're manipulated by the Nikon software. Since you'd then be comparing apples to apples (i.e. digital scans to digital scans), you don't have to worry about what an analog paper curve actually means if you try to transpose it to the digital world.

The only meaningful thing I'd be inclined to take away from the RA4 thing is that it's a steep, pronouncedly S-shaped curve. That, and saturation tends to be high (which I think is part of what you like, given the Egglestone reference).

PS: I think your thread works fine in this forum; I don't see any reason to move it - unless you'd like me to.

 

[albireo]

However...I consider this a red herring and I'd suggest just twisting some curves until you're happy with what you get. Ignore the whole RA4 paper thing. If you're scanning film, you're working with an abstraction of the actual negative, so you're never actually getting what the paper 'sees', and trying to approximate that will turn out to be a philosophical black hole and you'll run aground in metaphysical musings about the nature of reality and digital abstractions thereof. It's a dead-end street.

Koraks - thanks for the reply.

Sorry I don't see the red herring and -as stated- I'm not interested in twisting curves manually, but in understanding what makes this tool do what it does. I'm happy to continue using it, but I was curious about any first-hand insights on how it does it, and I wanted this info to be out there if available - as I suspect many other people out there might benefit from this and enjoy the results.

Having said that I thank you for your contribution and will read your blog with interest. Yes, a set of non-linear transformations (the 's-curve' you mention) is probably what's behind this - and it's perhaps similar to the non-linear transformations a black and white negative goes through when enlarged and wet printed (effect of lens flare+paper), eg

(from 'Way beyond Monochrome')

That, and saturation tends to be high (which I think is part of what you like, given the Egglestone reference).

I don't think it's a matter of saturation only, as stated in my post. I could easily achieve 'more saturation' by pushing a photoshop slider on any image, whether scanned or from a digital camera. There's more at play here. Perhaps selective saturation models based on visual-perceptual evaluation. Not sure yet.

 

[brbo]

I'd be more inclined that if you think Nikon got it "just" right in Nikon Scan (+ their scanner HW), it's pure coincidence.

If they knew they had something special developed and baked into Nikon Scan, I think at least someone in Nikon would come to the idea to use that in their digital cameras' jpg engines. Alas...

I wasn't too impressed by Nikon Scan, but I admit that I then didn't spend much time with it and also didn't wet print at the time. Now, I don't have any Nikon scanners anymore...

 

[Tom Kershaw]

I don't think it's a matter of saturation only, as stated in my post. I could easily achieve 'more saturation' by pushing a photoshop slider on any image, whether scanned or from a digital camera. There's more at play here. Perhaps selective saturation models based on visual-perceptual evaluation. Not sure yet.

I would tend to agree. While I have achieved good results with ColorPerfect from my Coolscan 9000 raw linear tiff file inputs, the ability of NikonScan to produce good results automatically does seem to indicate a distinct approach. I do intend to make some more practical evaluations of results in the future.

The colour in the Kodak Ektar example here is straight from NikonScan with an S-shape contrast curve applied in PhotoLine, but no other colour manipulation.

NikonScan Ektar sample

 

[albireo]

I'd be more inclined that if you think Nikon got it "just" right in Nikon Scan (+ their scanner HW), it's pure coincidence.

If they knew they had something special developed and baked into Nikon Scan, I think at least someone in Nikon would come to the idea to use that in their digital cameras' jpg engines. Alas...

Just in passing - I'm not suggesting Nikon is the only company that might have attempted to optimise scanning output via careful research+tweaking of HW+SW parameters against a perceptual endpoint.

I'm only focusing on Nikonscan here because that's what I've tried. It might as well be that Imacon machines+driver+SW did the same? Or Creo or Tango? Or any of those fancy specialised flatbed makers? Not sure, as I've not tried any of those.

Amongst the options I've tried given the hardware I have, Nikonscan is by far the best imo, and I'm trying to understand why.

 

[albireo]

I would tend to agree. While I have achieved good results with ColorPerfect from my Coolscan 9000 raw linear tiff file inputs, the ability of NikonScan to produce good results automatically does seem to indicate a distinct approach. I do intend to make some more practical evaluations of results in the future.

Thank you Tom, I'd be interested in your observations wrt this.

 

[Tom Kershaw]

• Vuescan Raw + Colourperfect plugin
• Vuescan Raw + NLP
• Vuescan Raw + Filmomat

Does ColorPerfect generally give you the best results out of these three? I sometimes find CP becomes confused by certain negatives, although doesn't overcorrect to neutrality like the Filmomat software.

 

[albireo]

Does ColourPerfect generally give you the best results out of these three?

Yes.

I sometimes find CP becomes confused by certain negatives, although doesn't overcorrect to neutrality like the Filmomat software.

Exactly my own findings.

 

[Tom Kershaw]

Here is a comparison. NikonScan on the right and ColorPerfect on the left. The CP version comes out of the plug-in with lower contrast so I've applied a stronger S-curve to approximately match the contrast with the NS version.

NikonScan / ColorPerfect comparison

 

[_T_]

The difference between these two is so minor as to be largely insignificant. You’d get more of a difference from looking at them on two different screens or by having them printed optically by two different technicians.

Also you say you’re color blind. So you can’t distinguish something like a third of the color information in a given image so if you say one way looks better than the other then that’s fine for your purposes but I wouldn’t use that as the basis for a hypothesis for use in further inquiry on the matter.

 

[radialMelt]

My issue with Nikonscan is handling of highlights/white point. In my use cases, it continually clips data, unless I bring down exposure significantly, which then clips the blacks.

You can see in that screenshot how much higher the white point is in the Nikonscan version.

 

[albireo]

The difference between these two is so minor as to be largely insignificant. You’d get more of a difference from looking at them on two different screens or by having them printed optically by two different technicians.

Also you say you’re color blind. So you can’t distinguish something like a third of the color information in a given image so if you say one way looks better than the other then that’s fine for your purposes but I wouldn’t use that as the basis for a hypothesis for use in further inquiry on the matter.

I'm sorry, who are you addressing this to? I haven't posted any examples. Also if you don't see differences between the two images posted by Tom I'd recommend an eye check (you sound way more colour blind than me) or a monitor check or both.

Apart from that seems like you have nothing to contribute towards my original point, so I'd move on to a more relevant thread.

 

[albireo]

My issue with Nikonscan is handling of highlights/white point. In my use cases, it continually clips data, unless I bring down exposure significantly, which then clips the blacks.

You can see in that screenshot how much higher the white point is in the Nikonscan version.

Agreed. This is the Nikonscan's Achilles Heel. Highlight clipping is what leads me to go for another method in case of extremely high contrast negatives. If you find workarounds for this, I'd be all ears. I haven't found a way, so I just fall back to Colourperfect for those cases (about 10% of my output IME).

 

[Tom Kershaw]

Absolutely agree. This is the Achille's Heel of Nikonscan. Highlight clipping is what leads me to go for another method in case of extremely high contrast negatives. If you have workarounds for this, I'd be all ears. I haven't found a way, so I just fall back to Colourperfect for those cases (about 10% of my output IME).

I have a suspicion that highlight and shadow clipping may have been what moved me towards ColorPerfect originally.

 

[Tom Kershaw]

Kodak Ektar comparison 2. NikonScan on the right and ColorPerfect on the left. The CP version comes out of the plug-in with lower contrast so I've applied a stronger S-curve to approximately match the contrast with the NS version.

NikonScan / ColorPerfect comparison 2

 

[koraks]

Interesting to see that the color balance goes in the opposite direction in both examples. In the first example, the color balance of the CP version is more yellow, in the second example, it's more blue/cyan. There is also a distinct difference in curve shape, with NikonScan emphasizing esp. mid-tone contrast more in both images and also exhibiting higher saturation.

 

[Tom Kershaw]

There is also a distinct difference in curve shape, with NikonScan emphasizing esp. mid-tone contrast more in both images and also exhibiting higher saturation.

I will try and put up some other comparisons as well. Ektar is not my generally used colour neg film, but I know these negatives, and have printed them on Kodak Supra Endura with good results, when that paper was fresh.

 

[Scott J.]

The ability of Nikon Scan to perform pretty good color inversions out of the box has been my experience, too. This has also been my experience with Flexcolor, the software that runs the Imacon/Hasselblad scanners. I've never been impressed by native color inversions produced by Vuescan or Silverfast. Even raw-scanning with Vuescan and SIlverfast (and inverting in Negative Lab Pro) gives pretty disappointing results to my eyes. I think the reality is that Nikon and Hasselblad were photographic companies that had a lot of expertise in the areas of photo technology and color reproduction. By comparison, the aftermarket vendors seem to be primarily software companies trying to fill a void (i.e., producing modern software capable of driving an old scanner on a current operating system), but without the same depth of knowledge regarding photo aesthetics and color. Nate at Negative Lab Pro appears to be an exception, in that he genuinely seems interested in photography and improving his product. I've not used ColorPerfect, but based on its supporters, its makers seem similarly committed.

One thing that I found with both Nikon Scan and Flexcolor, however, was that, even though they both did pretty well at inverting color negatives with just some auto settings applied, I could still get observably better color inversions scanning as positives (.TIF) and inverting in NLP. The key factor is that you have to turn off color management in both software packages (i.e., turn off all embedding and converting of the .TIFs to an .ICC profile). This "trick" doesn't seem to get stressed enough in the NLP forums, presumably because most people are scanning in some kind of raw format (which I'm assuming doesn't embed an .ICC profile?). Once I discovered that trick, my color inversions in NLP (acquired through Nikon Scan and Flexcolor) improved dramatically.

If you haven't tried doing this in Nikon Scan, it's definitely worth experimenting with and revisiting in NLP.

 

[albireo]

One thing that I found with both Nikon Scan and Flexcolor, however, was that, even though they both did pretty well at inverting color negatives with just some auto settings applied, I could still get observably better color inversions scanning as positives (.TIF) and inverting in NLP. The key factor is that you have to turn off color management in both software packages (i.e., turn off all embedding and converting of the .TIFs to an .ICC profile). This "trick" doesn't seem to get stressed enough in the NLP forums, presumably because most people are scanning in some kind of raw format (which I'm assuming doesn't embed an .ICC profile?). Once I discovered that trick, my color inversions in NLP (acquired through Nikon Scan and Flexcolor) improved dramatically.

If you haven't tried doing this in Nikon Scan, it's definitely worth experimenting with and revisiting in NLP.

Thank you Scott. I have to admit I haven't experimented much with NLP. I've only used it 'by the book' following the instructions here

Film Scanning Best Practices | Negative Lab Pro

Learn the settings and techniques for getting the best scans to work with in Negative Lab Pro

www.negativelabpro.com

So yes - as stated in my original post, my input to all third party tools I've tried is a raw linear 16bit/channel positive from Vuescan.

Based on the usage instructions above, and based my negatives, NLP returns results that are inferior (subjectively inferior of course - other people may of course prefer these) to Colourperfect. These are in turn less to my liking than the simple, all auto results I'm getting from Nikonscan. The only exception is with really high contrast, full sun images, where Nikonscan clips my highlights and Colourperfect has the edge.

 

[Les Sarile]

Prior to acquiring my Coolscan 5000 in early 2000, I had bought or used many desktop as well as all the minilab scanners which were all disappointing. My last test was the PMA show in Las Vegas and brought with me a Kodak Gold strip to be scanned on the Coolscan 5000, Minolta Dimage 5400 and a high end Durst system. After acquiring the 5000, I then sent out for 20" X 30" optical prints a sample from each of the color negatives and slides I used to compare the color, contrast and detail. I was very new to film then and didn't understand why Fuji Professional would take an interneg of the color slides in order to make the poster size print. I didn't really appreciate how well the straight up fully automatic results from the Coolscan + Nikonscan compares to these optical prints at that time. Over 20 years later - and tens of thousands of scans and many comparisons since, I have a great appreciation of how good the results are. I agree with the new observations that the Coolscan + Nikonscan combination is very hard to beat.

This has also left me wondering if the Nikon D850's built-in color negative conversion to be as good as Nikonscan?

BTW, I have since acquired the Coolscan 9000 and V and continue to run Nikonscan on a Windows Vista platform and the color, contrast and detail results are all exactly the same. It is when using ICE on Kodachrome or severely disfigured film that the 9000 separates itself. As well of course with being able to scan different film formats.

 

[Steven Lee]

I had experimented with a number of methods:

• Vuescan
• Vuescan 'advanced workflow'
• Vuescan Raw + Colourperfect plugin
• Vuescan Raw + NLP
• Vuescan Raw + Filmomat

Would be nice to see your results with all these methods! Otherwise it's hard to comment on your conclusions, except perhaps to point out that all of these methods have one common denominator - Vuescan. Maybe it introduces something undesirable into the scans? I have never used Vuescan myself. Back when I had my Coolscan 5000 I used the Nikon software. This was 15-20 years ago, but I remember not being perfectly happy with the inversions, so much so that I switched to shooting slides and eventually to 100% digital.

 

[albireo]

Would be nice to see your results with all these methods! Otherwise it's hard to comment on your conclusions, except perhaps to point out that all of these methods have one common denominator - Vuescan. Maybe it introduces something undesirable into the scans? I have never used Vuescan myself.

That's a good point actually. I can't exclude it.

I think I could probably extract a linear raw from Nikonscan too. I'll try feeding that into the third party tools and compare the results. Lacking that the only other tool left is probably Silverscan but I don't have a license.

I'll post some samples as soon as I can, though I don't know if this would add much. The thing is I'm not trying to convince anyone to use Nikonscan and there's definitely no "absolute" best.

I just find the Nikonscan results clearly more to my liking (importantly, more similar to my memory of the scene) and was surprised by this, given how old the tool is. Therefore I was interested in any other users sharing any relevant insights I might have missed on Nikon's inversion algorithms.

 

[Tom Kershaw]

Fujicolor Pro 160S. NikonScan on the right and ColorPerfect on the left.

NikonScan / ColorPerfect comparison 3

 

[albireo]

Example 1 - Kodak Colorplus 200 35mm --------------------------------------------------------------

1)

2)

3)

4)

Example 2 - Kodak Gold 200 35mm --------------------------------------------------------------

1)

2)

3)

4)

Example 3 - Fujicolor C200 made in Japan 35mm--------------------------------------------------------------

1)

2)

3)

4)

In no particular order: One is from a raw positive scanned in Vuescan v.9.8.26 and inverted in Colorperfect 2.25. One is the direct output of Nikonscan 4.0.3. One is from a raw positive scanned in Vuescan v.9.8.26 and inverted in Filmomat 1.35. One is from a raw positive scanned in Vuescan v.9.8.26 and inverted in Vuescan.

I've set a timer to make sure all of the above were ready in max 1 min Photoshop activity starting from loading the raw (or tiff in the case of Nikonscan and Filmomat) to final small jpeg generation. No custom color correction, no edits apart from: crop, set black point to histogram tail, resize, save to jpg.

Note for the tinkerers: I know you can improve any of these with a wise combo of PS color edits and/ord autocolor and/or curves. This is not the point of my exercise. The point of this is to see which method is giving me the most pleasing results out of the box or gets me closest to my memory of the scene, or gives me the output that is simpler to tweak into a final image.

Two questions
-what's your ranking, from best to worst?
-which tool produced which image?

Each image can be referenced by the number directly above it.