Help: Blue "noise"(?) in narrowband RGB scans. Losing my mind here.

[oCorvus]

I recently got a JackW Big Scanlight and have been struggling with small blue noise/specs which are especially visible in mid/dark neutrals.

I've attached some zoomed in images of the issue to this post (unable to attach full res due to file limits), and I have linked a Google Drive folder with the full res exports and Raw files. Additionally I also included multiple tests including trichromes, wideband white light scan (CS-LITE), exposure tests, and RGB level tests.

I think I am using the scanlight correctly with NegPy. I used a clear piece of developed film to calibrate the RGB light levels. I found that R255 G255 B100 was what got me a perfectly neutral film base on Portra 800 with each channel having equal level in the histogram.

I also did the Single-Shot Narrowband calibration in NegPy by taking individual photos of each of the naked light source and inputting them into NegPy's calibration module.

Things I've tried with no affect:

1. Adjusting scan exposure
2. Tried all the different demosaicing algorithms in NegPy
3. Sequential trichrome scan
4. Max RGB output (255 255 255) to check if blue intensity was a factor.

Scanning setup:

• Sony A7R4
• Sigma 105mm f2.8 ART Macro
• JackW Big Scanlight
• NegPy 0.62.0
• Linear RAW and Narrowband profile enabled in NegPy
• All example scans are 135 Portra 800

Is it possible that these are actually present on the film itself? As in these are legitimately noise from the dye clouds?

Google Drive Folder: https://drive.google.com/drive/folders/1RZtENSNZFjUQcWrypNgWw2p5cTCd8gU7?usp=sharing

[radioproektor]

NegPy can do strange things because it is automatic. Here is explanation in more predicted and simple fully manual workflow in PhotoLine image editor:

Problem 1. All your images are scanned with too low exposure. This could not provide enough bit depth data, especially in Blue channel if photos where scanned using white light. This problem also could be amplified because NegPy converts data from Linear to Log which decreases bit depth even more.

Add +1.5 stops in your camera and scan will be fine. Camera meter probably will show clipping, because it monitors inside smaller color space, but RAW data will remain unclipped.

Problem 2. This specific image was exposed incorrectly (overexposed) during shooting or during incorrect film development, so it require additional very special correction to bring it back to normal look:

In addition to normal Gamma adjustment it require large amount of Lightness (or Luminance) decrease:

Here is 100% crop. I can't see any blue noise artifacts in mage processed manually in Linear gamma.

[radioproektor]

I also looked at your other folders. Exp+2 is too high for sure. Camera sensor start to create magenta highlights artifacts which will affect film color. This will be a problem even if you remove that magenta from highlights in some other RAW editor and save as TIFF. So don't overexpose like this.
And by the way, in my examples i used DCB interpolation. I didn't add any additional noise reduction.

[oCorvus]

Thanks for testing the RAWs and showing your manual PhotoLine workflow.

My initial thought was also poor blue channel SNR, which is why I ran the blue brightness and exposure tests. Pushing capture exposure (+1 stop) still didn't fix the blue noise in NegPy (and I agree +2 stops was just an extreme test that obviously clipped).

I also found that manually inverting in PS did not seem to present the blue noise. I would ideally like to avoid a manual workflow given the volume of film that I shoot.

[radioproektor]

Well, any automatic tool will give you strange results if film is not perfect. And in the end you still spend time to adjusting that automatic tool as well.
Try to increase Black Point in Roll->Metering. This seems may fix the problem. Set DCB interpolation in RAW settings. It is slow, but produce very accurate moire-free image.

[oCorvus]

Thanks for the tip but that doesn't really fix the issue at all. It's just making it less pronounced by reducing contrast while also washing out the blacks of the image.

[radioproektor]

Well, i also tested it without calibration and ,if ignore uncalibrated color, tonally it looks near same as my manual correction and there is no any noise. So another possible problem - you build incorrect calibration which clips some channel. Or it is just another bug in color management which should be fixed.

[oCorvus]

I did include the images I used for RGB calibration and they aren't clipping.

Although I am also suspicious that it is a problem with the channel separation.

EDIT: although its presence in the trichrome scans doesn't really make sense then if its a calibration issue

[radioproektor]

ETTR version reported as "Clipped" during calibration and profile looks incorrect (too saturated and too contrast) and so blue noise is too visible.

1/60th version profile looks ok.

[oCorvus]

ok I definitely submitted the wrong file I guess for the ETTR because I did not get the clipping notification when I created the profile.

All of the sample exports I provided were using the 1/60th profile.

[radioproektor]

Green Light is clipped for sure. It is easy to see if compare histograms:

[oCorvus]

Yeah like I said I must have dragged the wrong file onto the drive folder.

The examples were using the 1/60th images anyways so I just removed the other folder.

[radioproektor]

1/60th calibrated profile is also rather saturated and so if increase contrast, it will amplify some amount of noise. It could be sensor noise due low camera exposure or some film grain particles remained.

[oCorvus]

If it is sensor noise it's not due to under exposure of the scan.

In the other example image even the +1 exposure (which is right below clipping) still presents the blue noise.

[radioproektor]

Just noticed that your files are 61MP. At this resolution it is not a sensor noise for sure, because sensor noise is way smaller in size. Sensor noise usually could be near same size as film grain on 24MP sensors or lower.

[oCorvus]

I am starting to wonder if it is an issue related to how the bayer filter on my camera reacts to the narrowband light.

As in it's due to the narrowband lights lack of smooth color transitions combined with the lack of color resolution inherent to bayer filters.

At 61 megapixels, only 15 megapixels of that are blue pixels.

Tonight I'm going to try a pixel shift scan to see if getting full color resolution solves it.

[radioproektor]

I didn't read NegPy manual, but just now realized that there are 3 stages in color. So i am a little bit confused if i understand it right.

1. Click "Linear RAW" button and create RGB calibration. (image start to look ok, but slightly oversaturated)
2. Click to "Narrowband" button. This will apply default LibRaw camera profile (or custom input ICC profile if it is set in preferences). Is it works in pair with calibration? This decreases saturation because LibRaw camera profiles are usually low saturated.
3. Apply Generic C41 correction to remove crosstalk increase saturation.

[oCorvus]

Yes I believe that is the correct method. From my understanding I can actually use the film stock specific matrices given I am using a calibrated narrowband setup.

I think NegPy, like many other inversion software, adds some compensation by default for the crosstalk of wideband scanning. I believe enabling the Narrowband setting disables that compensation.

[radioproektor]

So i guess this is how it works:
"Linear RAW" button bypasses camera WB and sets WB to 1-1-1. (Why don't name it somehow more logical? It sets WB, it have nothing with Linear gamma at all)
"Single Shot Narrowband RGB Calibration" adjusts sensor to R/G/B LEDs (This could be ignored If used White LED)
"Narrowband" button (very unintuitive name) apply LibRaw camera input ICC profile.
"Apply Generic C41 correction" attempts to remove remaining crosstalk color cast.

[radioproektor]

Also i still can't understand how to use Broadband White LED scans if i don't have custom made ICC profile. If set everything as recommended by setup (lamp icon) it simply assigns AdobeRGB ICC profile as input profile and so colors are incorrect. NegPy seems can't use default camera LibRaw profile in this scenario.
But if set custom camera ICC profile in "Export" tab settings - it works well.

[radioproektor]

https://github.com/marcinz606/NegPy/blob/main/docs/USER_GUIDE.md#103-calibration-what-your-rig-does-to-the-colors