Tri-Color Narrow Band RGB DSLR scanning

Greetings all, I have recently switched to using a narrow band RBG light for DSLR scanning. So far I have only done single capture with much improved results over 95 CRI, but I would like to utilize the capability of capturing individual RGB frames and compositing them so as to forgo use of any conversion software and get more "accurate" conversions. So far I have not had much success finding a step-by-step workflow that works. The closest would be this post by Alexi Machsas. It includes some confusing aspects of exporting linear TIFFs from RawTherapee that I don't quite understand so I skipped that step just to see what would happen.

I attempted following his other steps using just the raw files in PS, but did not have much success. Here are the steps outlined at the end of his rather lengthy piece found here:

https://medium.com/@alexi.maschas/color-negative-film-color-spaces-786e1d9903a4

In order to eliminate any gamma curve your RAW editor may apply to the image, you want to export a TIFF with “linear” gamma. The best way I’ve found to do this is to use RawTherapee and apply a “neutral” profile. If you’re comfortable with the command line, you can use dcraw to apply linear gamma with dcraw -g 0 0. You only want to adjust the gamma curve of the image after you’ve composited it, to account for the inherently low contrast in film negatives.

1. Create a new document and set the background color to black.
2. Paste each color channel image into your new document.
3. Set the blend mode of each color channel to “lighten” in the layers panel.
4. Add an “invert” adjustment layer.
5. Add a curves adjustment layer and select auto-options -> snap neutral midtones.
6. Add a channel mixer adjustment layer and reduce the amount of green and blue in the image until it looks correct. You can use an image with a grey card to calibrate these amounts.

For example, when I set the three layers to "lighten" all the layers disappear. Then when I add the "invert" adjustment layer the negatives reappear looking similar to how they would look when shot in single capture with the RGB light. I then added the curves adjustment layer, but this is where I am stuck. The negative is exactly that, still a negative. If I add an additional "invert" adjustment layer I then get a positive, but nowhere near a desirable conversion.

I realize this is rather niche, but surely there must be a way to layer and convert these properly. Does anyone have a workflow for this relatively new technique?

So what you're having trouble with is after RawTherapee, in another image program, combining the channels? If that's correct I think the guide to making trichromes out of 3 B&W images would be the steps you need. There's more than one way to do it.

Yes, I skipped the linear TIFF conversion in RawTherapee and worked off the raws in PS.

I am also not working with a debayered B/W sensor. I realize this is not technically correct and negates some of the benefit of trichromatic scanning but I am simply interested in understanding what kind of results I can get from this technique with the equipment on hand.

For Gimp: http://archive.brhfl.com/ph.brhfl.com/2012/09/07/gimp-trichromes/

Retro said:

This is how to do it in Photoshop (as described by Scott Bilotta in his Yahoo History of Photography Group):

The "goal is to create an image where the red record is copied to the red channel, the green to green and blue to blue. An easy way to do this in Photoshop is with the Merge Channels command. Merge Channels will take 3 separate same-size, flattened grayscale images and create a new RGB image. The command copies each image to one of the RGB channels. The result is a full-color photograph."

Click to expand...

originalwinslow said:

If I add an additional "invert" adjustment layer I then get a positive, but nowhere near a desirable conversion.

Click to expand...

There's nothing in the process that somehow inherently locks the color balance in what it should be. In other words: you'll still have to adjust the curve for each color channel separately. This is also true if you use a linear gamma (i.e. eliminate any S-curve applied by the raw converter). Don't stare yourself blind at that 'linear gamma' issue. It's not a fix for color balance, which still needs to be arrived at in another way, most likely manually unless you're working with some kind of color target that allows some form of calibration.

I'm not overly hung up on the linear gamma TIFF aspect, but I thought it would be worth mentioning in the off chance that was the root of my problem here.

Adjusting the individual color channels isn't giving me anything usable, at least not with this pipeline. That being said I'm not at all an expert with this type of tinkering and clearly doing something wrong.

I don't think using GIMP is going to be a viable workflow for me. The author themselves state it's an outdated and frustrating software...and that was over 14 years ago.

originalwinslow said:

The author themselves state it's an outdated and frustrating software...and that was over 14 years ago.

Click to expand...

I use it all the time. The latest release was...idk, a week ago or so? The problem is not GIMP. Do the same thing in Photoshop and you'll notice it's exactly just as challenging.

The problem is that color neg inversions, esp. done manually, just takes experience and no software substitutes for that. You can get quicker results with the well-known automated tools, but they all have the same issue: consistency (or lack thereof). However, in many cases, that's not a very hard requirement, which I think explains why many people prefer that approach.

Any chance you could upload the TIFFs or even RAWs you've made of this test image so I can give it a spin? If that works, I can explain what I did and perhaps that helps any.

Sure, that would be great! I can post a download link for the raws.

https://originalwinslow.fromsmash.com/TRICOLOR-EXAMPLES

I don't mean to bash GIMP, working with a new set of tools is always daunting.

For reference, here is a single capture NLP conversion from the narrow band RGB with no edits applied.

A lot depends on taste of course.

What I did:
* Convert each NEF to TIFF 16 bit. I turned off the highlight recovery adjustment in RawTherapee but didn't touch anything else (OK I did rotate 180 deg.)
* Import all TIFFs into a single image as layers in GIMP
* Adjusted curves on each layer with a simple linear adjustment so that the histogram for each layer occupied the entire width (so set black & white points basically)
* Then converted each layer to monochrome using the Mono Mixer. I only used the blue channel for the blue layer, green for green and red for red (so left the other colors at zero).
* Switched to greyscale mode
* Used the Colors -> Components -> Compose function to assemble the layers into a single RGB image
At this point I had basically the same as what you'd get with a normal color negative scan before it's inverted/made positive.
So for the final conversion I added an inversion (linear curve from top left to bottom right). Then I added separate color adjustment curves for each of the colors. I looked for the shadow patch at the left side of the frame (the little cave/shadow between the rocks at the image frame) and the brightest cloud just above the mountains as black resp. white points. The rest of the color balancing was basically visual and to taste. This shows the final adjustment:

Note how each color channel has its own S-curved adjustment. Note also that I pushed the blue curve pretty far to the right.

The bottom left corner of the image remains a little weird; I think this is a scanning artifact as it seems to have received a little less exposure during DSLR scanning. Could this be the case?

I kept the final balance much cooler than yours; this is because most of the scene is in the shade of the mountains and I personally prefer to keep shaded areas cooler as that's how they realistically look as well. Moreover, it benefits the green foliage IMO to not push towards yellow too much, which loses the richness in the greens very easily. However, this is all a matter of taste more than 'right' or 'wrong'.

Furthermore I allowed the shadows to drop down pretty far along the lower part of the image; these could easily be lifted (note the S-curved adjustment on the Value channel; the lower part could be lifted if desired). Again; a matter of preference; the negative itself and the scan have differentiation there that can be used to fill in those shadow areas as your NLP conversion also shows.

Thanks for the in-depth explanation koraks, much appreciated. I enjoy seeing how your subjective and objective interpretations vary from eye to eye. I think it very much emphasizes that there is no "correct" conversion, only pleasing ones. This was shot on 500T developed in my DIY ECN-2.

I will give this a shot in GIMP. All in all, it seems doing manual conversions is going to be far too time consumptive to make it particularly viable but it's something extra to have in the back pocket.

As far as the anomaly in the corner goes...it is the one thing that has confused me about this RGB light. It creates a strong vignette that also has a tint shift. It is also stronger on one side of the frame. That is what you're seeing with the red in the corner. This does not occur with white light. I had thought that converting individual RGB frames would perhaps eliminate this, but that is not the case. It is not caused by the film holder. I have been able to eliminate this by filling less of the frame with the negative and creating a custom radial mask with tint adjustments or doing FFC. I am stumped as to what is causing it. I thought about creating a separate thread to discuss if others have run into this issue.

Here is an exaggerated example. This is a shot of just the light in RGB mode, no holder present, with contrast pumped up. I am using the big scanlight by jackw01. I asked him about it and he didn't have much of an answer. I shot examples with three different cameras and lenses and always got similar (but not identical) results.

Ah, that's an interesting effect. Could be related to angle of the light as seen through the diffuser, maybe? I'd have to think about it. I think the Scanlight device is fairly thin which means it allows for little light mixing and diffusion. Maybe with a larger diffusion chamber the problem would go away. IDK, just a hunch.

You're definitely right about the subjective/objective and no right & wrong! Also, I did this in one particular way and order, but there are many variations to the theme that should also work.

The big scanlight is actually quite deep compared to other lights I have seen. 5cm roughly. I believe it was designed this way specifically to eliminate poor light mixing. Jack said that it could maybe be caused by IR light reflecting off the lens or holder, but when taking these sample images I had the camera well away from the light source such that any reflection would be negligible.

Maybe it is a faulty device. One other user on the NLP forum told me they saw a similar effect which was reduced by lifting the film plane farther from the surface of the light. I purchased some NR glass to elevate the holder a bit, but it has yet to arrive.

It's not just vignetting on your taking lens? Because that would look pretty similar. Only the color aberrations are a little odd.

If I was to compose the images, I would use Darktable, applying a a "none" workflow to all images and WB as D50 and ColorCal as bypass so they stay linear. Then I whould open one, and us the "composite" module to add the other two layers (that can still be edited independently), after everything is done I whould create a 2nd instance for Color Calibration to correct source WB, and use negadoctor module for removal of orange base, wb of the film, and tonal mapping/correction and exposure of the image. You can create a preset to do this automatically.
If I was to build the exposure unit from scratch, I would make it to be able to quickly flash all three lights in a row in a single shot of the camera, being able to control the individual timing. That would eliminate much of the future processing

Chromium VI said:

I would make it to be able to quickly flash all three lights in a row in a single shot of the camera, being able to control the individual timing.

Click to expand...

But that would defeat the purpose of the 'quasi-monochrome' approach that @originalwinslow tried out here. I do agree that it would make processing easier, and your approach would at the same time allow the channels to be balanced so as to hit the sweet spot of the sensor's dynamic range across all three channels. Btw, this could also be achieved with PWM dimming.

originalwinslow said:

As far as the anomaly in the corner goes...it is the one thing that has confused me about this RGB light. It creates a strong vignette that also has a tint shift. It is also stronger on one side of the frame. That is what you're seeing with the red in the corner. This does not occur with white light. I had thought that converting individual RGB frames would perhaps eliminate this, but that is not the case. It is not caused by the film holder. I have been able to eliminate this by filling less of the frame with the negative and creating a custom radial mask with tint adjustments or doing FFC. I am stumped as to what is causing it. I thought about creating a separate thread to discuss if others have run into this issue.

Here is an exaggerated example. This is a shot of just the light in RGB mode, no holder present, with contrast pumped up. I am using the big scanlight by jackw01. I asked him about it and he didn't have much of an answer. I shot examples with three different cameras and lenses and always got similar (but not identical) results.

Click to expand...

So I built the smaller scanlight and frankly even the v3 diffusion setup leaves a lot to be desired for color and intensity evenness. I ended up modifying the diffuser housing design and stretching it approximately 20% taller, printing it in white with a textured inside, and adding a spacer and second diffuser to the top. Finally, that got it sufficiently even for use on 35mm. I don't believe I've tried medium format on it yet.

koraks said:

But that would defeat the purpose of the 'quasi-monochrome' approach that @originalwinslow tried out here. I do agree that it would make processing easier, and your approach would at the same time allow the channels to be balanced so as to hit the sweet spot of the sensor's dynamic range across all three channels. Btw, this could also be achieved with PWM dimming.

Click to expand...

If narrow band leds are used, that would lead to almost identical results, perhaps a beter noise on the 3shot version though... One thing that would be interesting was trying the 3shot aproach with a camera with the bayer screen removed, and using a demosaic option (like one available on darktable) that would allow to extract 4x the resolution of the sensor with such mod.

koraks said:

It's not just vignetting on your taking lens? Because that would look pretty similar. Only the color aberrations are a little odd.

Click to expand...

This is a much stronger vignette than I would expect to see from a quality macro lens stopped down to f8. Shots of the 95 CRI white mode do not exhibit this level of vignetting either.

thinkbrown said:

So I built the smaller scanlight and frankly even the v3 diffusion setup leaves a lot to be desired for color and intensity evenness. I ended up modifying the diffuser housing design and stretching it approximately 20% taller, printing it in white with a textured inside, and adding a spacer and second diffuser to the top. Finally, that got it sufficiently even for use on 35mm. I don't believe I've tried medium format on it yet.

Click to expand...

The big scanlight is a very different design with a larger housing similar to what you describe. Consistency across the panel is much improved as a result. At least that's the claim.

I don’t have my Scanlight yet (ordered), but here is my general manual conversion method. In most cases, it can match or beat NLP.

- Assign linear profile in ACR. Linear profile is created in Adobe DNG editor by converting a Sony ARW file to DNG once and then importing. Change the curve to Linear to turn off Adobe’s built in curve, then hit export. This only needs to be done once and it will then appear in LR and ACR as an option for any raw format going forward.

- In ACR, white balance off blank film area and open file in PS

- Create Curves layer and manually invert each individual RGB channel - not the composite RGB channel. This can be saved a preset.

- Set white and black points in each individual RGB channel (not composite) using the Show Clipping option to get precise. This usually gets the color 99% of the way there.

- Add more Curves/Color Balance/Selective Color layers to adjust color and contrast to taste.

@silvergelatin note that the additional 'complication' here is the use of three separate 'scans', one for each color channel, that need to be merged.

koraks said:

@silvergelatin note that the additional 'complication' here is the use of three separate 'scans', one for each color channel, that need to be merged.

Click to expand...

oops

originalwinslow said:

One other user on the NLP forum told me they saw a similar effect which was reduced by lifting the film plane farther from the surface of the light.

Click to expand...

Absolutely, I can get much worse results from what looks like vignetting (but mostly isn't) based on how far the film plane is from my CS-Lite. I've recently had to increase the distance and it improved things a lot. I tried adding extra masking, but the distance increase improved it more than the masking did.

Here is an example of the vignette before and after raising the film. The difference was maybe two centimeters at the most. Not apples-to-apples exactly, as the before is a 6x9 and the after is a 6x4.5, and the before is B&W and the after is color, but I've adjusted the curves for both based on how I adjusted the curves for images on each roll, so you can see how much stronger the "vignette" was happening before the film was raised. These are both blank frames on the film. One of the ways to account for this is to take a pic of a blank frame on each roll and layer it with the Divide blending mode, but it's better if you don't have to (and sometimes you won't have a blank frame available).

loccdor said:

View attachment 419311

Click to expand...

This looks rather similar to the issue I'm having. The NR glass arrives in 10 days and I can test it out.

koraks said:

@silvergelatin note that the additional 'complication' here is the use of three separate 'scans', one for each color channel, that need to be merged.

Click to expand...

I imagine it's only a matter of time before someone designs a dedicated standalone software that could perform this action automatically. That coupled with a true b/w sensor could make frontier/noritsu scanners obsolete with the additional upside of modern curves controls.

originalwinslow said:

I imagine it's only a matter of time before someone designs a dedicated standalone software that could perform this action automatically. That coupled with a true b/w sensor could make frontier/noritsu scanners obsolete with the additional upside of modern curves controls.

Click to expand...

The new NegPy scanning software, splits the channels and works them as monochrome for correcting negatives color, it's still on development but the color correction works very promising, it whoul be great if the team could add such option (to use three bw shots).

originalwinslow said:

I imagine it's only a matter of time before someone designs a dedicated standalone software that could perform this action automatically. That coupled with a true b/w sensor could make frontier/noritsu scanners obsolete with the additional upside of modern curves controls.

Click to expand...

Breaking apart the old Kodak RFS 3570's software might give you the answers - from recall, it used RGB filters and a monochrome CCD. It is not a hugely complex undertaking compared to de-Bayering.