Flatfield correction leaves dark residual gradient even when self-correcting reference frame.

[oCorvus]

Been running into an issue where flatfield correction seems to add a dark gradient to the bottom of my images.

I take a flatfield reference image after every roll scanned. I simply take a photo of my empty film carrier with aperture and focus kept exactly the same as scanning. Only increasing the shutter speed to avoid clipping the naked light source.

I decided to import the flat field calibration image into NegPy as a scan and use the same image to correct itself. I assumed that if I corrected an image of my lightsource with itself as the flatfield reference, that it would result in a perfectly uniform frame.

To my surprise it did not. The exact dark gradient I've been seeing in my images suddenly appears.

Here is a flat field reference image imported into NegPy. NegPy applies it's inversion and you can see the falloff towards the corners.

Now here is the same image but with flat field correction applied using itself as the reference image. It fixes the corner fall off but adds this dark gradient at the bottom of the frame.

Any ideas on what might be causing this? Seems to be a thing with all of my flat field images.

Gear:

• Sony A7R4
• Sigma 105mm f2.8 Macro ART
• JackW Big Scanlight
• Negative Supply 35mm Pro Carrier

EDIT:

link to RAW file of flat field images: https://drive.google.com/drive/folders/1mCchvJFwckyNqTLZNAHUt8PTIDaV_Y2q?usp=sharing

[thetalkingdrum]

@oCorvus I made a possible fix, please try it out: #1315

It was a bit difficult to get the Negpy to show nothing but a black frame. I had to turn off Linear Raw and play around with the luma range clip, but it does seems to work. You have to build a new flat-field profile for this to work.

[radioproektor]

That NegPy correction looks really strange. Here is manual correction in PhotoLine. Blur+Levels+Divide blending mode. Quick and simple. It should look like perfect uniform gray area and balck frame also should be 99.9% removed.
In normal situation FFC should be white balanced to keep unchanged WB of source image, but in this example i skip WB because non-wb FFC fixes non-wb source color cast:

[radioproektor]

I guess problem could be because NegPy apply too many Blur to FFC frame. In my workflow i use Blur value 5 (not sure what it means exactly, radius or diameter in pixels or something else) Larger Blur values create same effect as you can see in NegPy - FFC don't filter small details and spots, and film frame edges also became unfiltered.

[radioproektor]

Here is example with different blur values. To make effect better visible i add additional layer of Levels on top and clip very large amount of black and white values.

[thetalkingdrum]

@radioproektor your Blur + Divide + Levels is a good way to look at it. My screenshot above was NegPy's inverted print, which squashes an even flat to near black, so it can't be compared with a divided image.

Here is DSC06533 corrected by itself, shown the way your Levels layer shows it (gray = even, ±5%):

Left is NegPy before #1315: the bright band at the bottom is the dark carrier edge bleeding into the blur, the same bright edges you get at blur 255. Right is with #1315, where the carrier is masked out of the blur and the result is within about 2%.

On blur size: a small blur also corrects sensor dust, but it needs a full-resolution gain map and copies anything that moved between the flat and the scan into every frame, so NegPy corrects only the smooth falloff.

Since the print can't show this, #1319 adds a Check Flat Field view that shows it in the app.

[radioproektor]

To be honest i don't use NegPy, but just test it for fun. I really don't understand what is gain map and why NegPy do simple FFC correction in so complex and incorrect way. So called "only smooth falloff" feels like is some sort of fake FFC that only makes things worse.

You just need to use find correct amount of Blur. Too small blur is bad. Too much Blur is also bad.

Normal FFC correction done like this:
Light source and camera positions should be permanently fixed. Lens focus and zoom also should be set once and fixed.
In Linear gamma:

1. FFC frame: WB -> Blur -> Levels (automatically clip white point to 100%)
2. Divide with source layer
3. Apply all other film neg corrections...

[thetalkingdrum]

@radioproektor The “gain map” is just your Blur + Levels result, inverted and saved once per profile, so it doesn’t have to be recomputed for every frame. Multiplying by it is the same as your Divide, done per channel, which also covers your WB step. So the math is the same as yours.

You’re right about blur size, though. NegPy’s 1/16 of the frame dates from the first version and was never tuned. I tested it on oCorvus’s two flats, shot two days apart: a gain from one corrects the other about 3× better with a blur around 1/60, with no added noise. I’ll open a PR for that.

One difference to keep: the carrier has to be masked out of the blur, or its black edge bleeds into the frame, which is what #1315 fixes. With a smaller blur, an unmasked edge gets worse, not better.

[radioproektor]

Yes, Invert+Multiply is the same as Divide