Color Calibration: Remove tint from ND filter

Software darktable
1

I took some pictures with my ND1000 filter, which creates some tint in the image.
I’m confident, that I can remove this tint with the color calibration module, but I’m not so sure how to do this actually…
So far, I found two options that work reasonable well:

  • use the AI autodetect features
  • do a manual WB

However, I also found some hints on the internet, that I should take two pictures of a neutral colored wall, one with the filter and one without, and then note the shift in WB and use it for all images. However, these tutorials use Color Temperature in Kelvin and Tint and not hue and chroma…
Nevertheless, I tried it, noted the shift in hue and chroma and applied it to one one image taken earlier and it does not look too bad!
What I do not understand though is how I should set the WB initially. I used “as shot in camera” and then switched to “custom”. This seemed to work, but is this reliable and always correct?
I have my doubts that this is the correct way to do it - even though it worked for this single instance :sweat_smile:

Another interesting question would be, how I could store this as a preset… Is there a feature in darktable that says “take the value that is currently set and add X”? I think that does not exists, right?

2

If you had a colorchecker and took a shot then you could likely correct by simply using the colorchecker correction in CC module… but that might not be an option for you

3

Hmmm… an ND-filter ought not to create a colour cast — at least
not those of high quality.

However, a swift method to find a “general” cast is to run the image through
The Gimp à la one of the usual WB-finding methods (there were two presented
in the forum the other day). The data found there can then be used in darktable.

Holler if you can’t find them, and I’ll dig up more info.

Have fun!
Claes in Lund, Sweden

1 Like
4

yes, and back then (2014) when I bought it, I took the one where the internet said it was of proper quality :smiley:
I have a “B+W ND 1000x F-Pro 110 SH”.
I did not use it for years, but now I noticed that it has actually a strong red-ish tone. Here are the two testshots side to side I took today of a white wall (the WB is the same in both cases):

image
image720×475 24.4 KB

Maybe it wasn’t the best filter after all…

Unfortunately, I do not have one… But I thought the same, it would certainly help here.

2 Likes
5

could you post an image with and without the filter being on so I could try and look for a simple solution. BTW, I bought a variable ND filter and at very strong settings I also experienced color shifts. It may be the strength of the ND filter coming into play here, but lets see how we can overcome that in DT.

2 Likes
6

Sure, here are the two images from yesterday:

IMG_6074.CR3 (14.9 MB) → No Filter
IMG_6075.CR3 (20.8 MB) → With Filter

Both are licensed as Public Domain.
I noticed, that the filtered image needs about 0.03EV more exposure to get to the same brightness.

I took them with different ISO to keep the shutter speed low - I might repeat that in brighter conditions…
I also read you should take a picture of a white screen on a D65 calibrated monitor and use that as a reference. I might try that as well.

edit: I thought about it and I think a better (not sure if correct though) way would be to add a second color calibration module and apply a correction to the RGB channels. With the given image it is relatively easy to do this, by moving the sliders such that in the color waveform all three lines are over each other.
However, I’m not sure if the wall I photographed is perfectly gray - I think I’ll repeat this with a calibrated monitor.
Nevertheless, I tested it that way now and applied the correction to an image and it also looks okay. I think the real test would be to take pictures with and without the filter of some scene and then apply the RGB correction.

the correction I found would be:

R +0.87
G +0.93
B +1.365
1 Like
7

Morning, @reox,

I have a feeling that your are a bit overthinking (could actually also be that I am underthinking) the problem.

You are using darktable, and dt has a suberb submodule for tasks such as this.
Look up spot color mapping within the color calibration module. darktable 4.4 user manual - color calibration

Have fun!
Claes in Lund, Sweden

8

I’m certainly overthinking this :smile:
As I said in the beginning, I managed to get reasonable WB just with the AI tool.
However, I saw that video where the guy said he measured the color shift once and applies this directly (in Adobe Bridge…)
I thought, that this would be neat to have a preset ready when images were shot with the filter on.

I already read this section regarding the spot calibration, but I cannot figure out how I can use this in my case. For this to work I would need to know a color spot in the photo, that matches some known reference photo.
I think it is easier to use flowing water (the main usecase for me) as a gray reference instead and do the WB on that spot. Or maybe I misunderstood how the spot calibration works :sweat_smile:

9

The comparison here is based upon using first the color spot mapping option, then the exposure mapping option, and then apply denoise (profiled). The results are similar but a slight difference is observed. Possibly doing the three modules I used in a different order may give a closer result. The fact that one is 800 ISO and the other 6400 ISO may also play into the equation.
NDfilter-comparision

10

Morning @reox!

No. It is much easier than that.
Try this skeleton approach in darktable:

  • open a (non-touched) ND image.
  • do what wizardry you like to make that image look non-ND.
  • in lighttable, mark the image you just worked on, goto styles → create.
  • name your style, for instance as fixND.

After that, it will mostly be sufficient to open up another ND image, and select style fixND.

I write “mostly” since you often will have to tweak this or that parameter to
make it look OK…

Have fun!
Claes in Lund, Sweden

1 Like
11

BTW, this is one of the reasons that instead of using 10-stop ND filters, I use a weaker (3-4 stop) ND filter to get the shutter time that does not cause highlight clipping long enough that the camera won’t fill up its buffer in continuous drive mode. Then I take a sequence of images and average-stack them to synthesize a much lower effective ISO. Has the benefit of much higher dynamic range since you’ve collected much more light.

2 Likes
12

One can create styles directly in the darkroom.

The button to the right of the “compress history stack” button allows you to create a new style from the history stack of the current image (darktable 4.6 user manual - history stack)

However, if one tweaks the primary white balance modules (either white balance or, if using the ‘modern’ approach, the first instance of color calibration), the result will not be applicable to other images shot under different lighting.

1 Like
13

You could try this old module…

I think there is a blog post on its use I will dig that up and add…

Works not bad for the images you provided…however might be less effective on a full photo… I think its meant more for artistic rather than correction

image
image1920×1080 265 KB

1 Like
14

I also thought about taking a series of images without the filter and then using a median stacking - to get from 1/250 to about 0.5s is not that hard. My camera can even do this in-camera for up to 9 photos.
Interestingly, also my Pixel 7 has a long exposure mode - even though you cannot really set a specific exposure time (at least I haven’t figured that out yet). I don’t know why they do not put something like this into the other cameras as well: That would be a really neat feature.
Anyways, for 58mm lens diameters I have a ND64 filter, which I think produces a blue color - but I think it is much less than the ND1000 (which I have only for 77mm)

15

What package do you find to be best for the average stacking?

16

For simple average stacking, siril is pretty clunky (since it requires converting every input file to FITS first), so I stopped using it and switched to a tool I wrote myself that is MUCH faster - pyimageconvert/pystack.py at master · Entropy512/pyimageconvert · GitHub

4 Likes
17

While it’s not possible to duplicate the white balance module for this correction, it’s indeed possible to duplicate the channel mixer module with this settings ( just an approximation)

RED
r +0.964
g + 0
b + 0
GREEN
r+0
g+1.22
b+0
BLU
r+0
g+0
b+1.410

IMG_6075.CR3.xmp (5.0 KB)
original

IMG_6074
IMG_60742048×1366 383 KB

White balanced with channel mixer

IMG_6075_01
IMG_6075_012048×1366 1.04 MB

3 Likes
18

Thanks, Andy, I’ll check it out.

19

I tested it now with the D65 calibrated screen and trying to get the RGB values from that. It seems to work and the coefficients are then R+0.979, G+1.016, B+1.27

However, I still cannot simply put this onto another image and get the correction… I’m not so sure if this actually works :smiley:
Nevertheless, using manual WB gets me very far, so using a profile would be nice but is not a must-have!

20

It seems to me that the root issue is that the filter produces an offset - i.e. you need a relative correction, which isn’t really doable as a preset in WB or color calibration to the best of my knowledge. Hence @age’s solution using channel mixer seems a good option.
Sorry, I guess this is stating the obvious - but I was thinking about it last night…