5
\$\begingroup\$

While editing some landscape shots with stars, I tried to use darkframes to reduce the noise.
More precisely, my approach was to take a series of shots, then firstly to subtract dark frames from each shot, secondly to use the mean of the series for the foreground to further reduce noise, and thirdly to use an astro stacking tool (Sequator) to stack the sky.

Instead of reducing noise, the darkframe subtraction:

  1. increased the noise- or rather, added some dark/monochrome noise.
  2. changed the white-balance/tinted the image.
    (see below)
    I do not understand why this is happening/What I am doing wrong.

Procedure/Employed Troubleshooting:

  • All photos were shot in succession, with the same settings (15sec, @ISO6400, in-camera dark frame disabled).

  • All photos were shot with the same white balance.

  • While shooting the darkframes, both the lens cap and the viewfinder cover were applied.

  • Photos were imported from my Pentax K1ii, converted to DNG in LR, and exported to PS without any editing/import presets applied.

  • I used PS, placed the darkframe layer(s) above my picture, and used the subtract blending mode.

  • I followed basic instructions found here/in various videos on dark frame subtraction in photoshop. Note that basically, all of those cover dark frame subtraction with one frame (or use tools other than photoshop). I have tried both using one, and 3 frames. The results are similar, albeit more pronounced with 3.

  • I used the free tool "sequator" to subtract darkframes instead (and to align stars). Adding the dark frames here made absolutely no difference.

  • (This is an edit/composite done with the frames I tried to subtract darkframes of)

  • A crop of the first picture, with (3) dark frames subtracted:
    with (3) dark frames

  • A crop of the second picture, without dark frames subtracted:
    without dark frames

\$\endgroup\$
14
  • \$\begingroup\$ Why is the in camera darkframe disabled? \$\endgroup\$ Apr 16, 2021 at 23:49
  • \$\begingroup\$ Why are three darkframes being subtracted? \$\endgroup\$ Apr 16, 2021 at 23:54
  • 1
    \$\begingroup\$ Did you feed Sequator the DNG images or raw images? You should be feeding Sequator raw images. Sequator uses dcraw to convert raw images to a format that doesn't have any corrections. \$\endgroup\$
    – qrk
    Apr 17, 2021 at 0:11
  • \$\begingroup\$ Why shoot PEF instead of in-camera DNG when you want DNG later in the workflow? \$\endgroup\$ Apr 17, 2021 at 2:41
  • \$\begingroup\$ @qrk I did feed uncompressed 16-Bit .tiffs exported from LR to Sequator. While I'm sort of happy with the result, here, I will definitely try if I get a different result with my .pef files \$\endgroup\$
    – Kaschmir
    Apr 17, 2021 at 9:59

4 Answers 4

5
\$\begingroup\$

You misunderstand the purpose of dark-frame-subtraction. While it is a technique used to reduce noise, it only reduces noise that is consistently output from the sensor. Any read noise due to the circuitry or uneven output such as hot-pixels. It does not reduce random noise.

When you used dark-frame-subtraction on an image with noise that is primarily random, you are in fact simply adding more noise. Some pixels may look better but overall, there will be more noise. Noise in your image and your dark frame is very high because you are shooting at ISO 6400. It is important to capture using the lowest ISO possible. If you cannot lengthen exposure, open the aperture and if you cannot either, then consider buying a brighter lens.

To reduce random noise using a multi-frame technique, the better approach is to average images. This means capturing the same subject multiple times and then blending the resulting images together, so that random noise gets averaged out. It will not disappear but the more shots you can average, the lower the final noise will be. Of course, anything that moves in the scene causes problems with this but if you are capturing stars within a short period of time, their relative position will be consistent and you can align the image-stack before averaging. If you have foreground elements with foliage, then this technique is not really usable.

\$\endgroup\$
3
  • 1
    \$\begingroup\$ Thanks, this was insightful regarding darkframes. \$\endgroup\$
    – Kaschmir
    Apr 18, 2021 at 19:04
  • \$\begingroup\$ I take away that in order to reduce hotpixels with darkframes (which are a huge issue in summer for any D800/K1ii user), I would need to average many dfs to reduce the noise that I otherwise add to the picture? By the way, I did average 8 shots for the foreground in the link above. More would have been better. And for star photography, there is no alternative to high ISO, unless you want to use a sky tracker or the sensor shift option of my Pentax. But then, you end up with composites that are a hassle to merge properly. \$\endgroup\$
    – Kaschmir
    Apr 18, 2021 at 19:13
  • \$\begingroup\$ Hot-pixels are usually not random at a given temperature, so that's what dark-frame subtraction is usually good at. It's all the other noise, most amplification noise that occurs when boosting ISO. \$\endgroup\$
    – Itai
    Apr 19, 2021 at 1:30
2
\$\begingroup\$

Instead of reducing noise, the darkframe subtraction increased the noise- or rather, added some dark/monochrome noise.

How long was your session? What was the ambient temperature? Was the camera at ambient temperature at the beginning of the session? At what point in the session did you take dark frames?

If the camera was at ambient temperature when you started, and the session was long enough that the sensor temperature rose significantly during the session, then you need to apply dark frames made periodically throughout the session that match frames taken when the sensor was near the same temperature.

  • If you apply dark frames taken when the sensor was much cooler to light frames taken when the sensor was much warmer, there will be increased noise caused by the higher temperature that won't be eliminated by the dark frame taken when the sensor was cooler.
  • If you apply dark frames taken when the sensor was much warmer to light frames taken when the sensor was much cooler, you will "eliminate" noise that wasn't there in the first place and parts of the image will be darker than the background luminance of the sky.

Instead of reducing noise, the darkframe subtraction changed the white-balance/tinted the image.

The color is different because without dark frame subtraction or any other noise reduction being applied, the predominant color of most astronomical photos will be the colors introduced by the chrominance noise in the image.

Photos were imported from my Pentax K1ii, converted to DNG in LR, and exported to PS without any editing/import presets applied.

When you imported to LR, your current default settings for LR would have been applied to the raw data. There's no such thing as an "unedited" raw photo that looks anything like a photo on your screen. If you're not telling the app how to interpret the linear monochrome luminance values contained in the raw data, you're letting the app decide on its own.

\$\endgroup\$
4
  • \$\begingroup\$ Thank you for the input. 1) The session was relatively short (first frame taken at 23:35:51, first dark frame at 23:41:18; last dark frame at 23:42:28). 2) Ambient temperature was below freezing (~-3°C). 3) Thanks for the clarification on the dark frames' effect on color temperature 4) Regarding the LR import/export: I realize I did apply some slight sharpening, here. That's it. I don't expect it to change much, but I could be wrong. \$\endgroup\$
    – Kaschmir
    Apr 16, 2021 at 19:53
  • \$\begingroup\$ Addendum on 2) No other shots than the 8 frames + 3 darkframes were taken in between. Moreover, I habitually minimize time spent in live view for that reason, and move the display as far away from the camera body as possible (not sure if that makes a difference, but I imagine it could let the sensor/body cool off more quickly). \$\endgroup\$
    – Kaschmir
    Apr 16, 2021 at 20:02
  • \$\begingroup\$ Sharpening has a direct influence on noise. In a way, NR and sharpening are like the numerator and denominator of a fraction that determines how noisy an image is. \$\endgroup\$
    – Michael C
    Apr 16, 2021 at 20:12
  • \$\begingroup\$ Thanks, I will try again tomorrow, and report back. I still don't think it can be the issue, though, since it was rather low \$\endgroup\$
    – Kaschmir
    Apr 16, 2021 at 20:14
2
\$\begingroup\$

A while back I tried some of these images and found that it was much more convenient to take a dark frame for a group of long exposures rather than sit and wait for a minute every time I took a minute long exposure (makes star trails look like morse code). At the time I was using RawTherapee which had a darkframe substraction module which worked really well, see RawPedia.
For various reasons I've switched to darktable and darkframe subtraction is something that I miss as it is not included. The work around which I found was to use Siril, a very powerful but technical and heavy going package more geared towards astronomers. As such it can undertake "Pre-processing of images with multi-channel offset, dark and flat images", "Image registration" and "Image stacking, with optional additive or multiplicative normalization". see Siril website.

I tried the photoshop (Gimp) route and found the results poor in comparison which was why I posted this question : How does in camera dark frame noise reduction work?.

\$\endgroup\$
1
\$\begingroup\$

Try turning on in-camera darkframe subtraction instead. The designers of the camera have insight into how to best mitigate noise because they have detailed knowledge of the sensor and the noise inherent in the system.

The in-camera noise reduction is part of the overall-engineering design specific to the camera. It is the product of informed expert engineering opinion.

Dark frame subtraction may not be necessary. Because of video based improvements, more recent cameras tend toward better thermal management.

Testing different methods by isolating one variable at a time has a long tradition in photography. Because what works for one photographer may not meet another’s expectations.

The process described by the question has a lot of “moving parts” — a lot of techniques to learn all at once and many steps that need to be done the same way in the same order to produce consistent results...consistent results are the first step toward achieving consistently good results.

So more generally, simplicity is a reason to try in-camera dark frame reduction or no dark frame reduction. Then the variables are all in the post processing. Once the post processing limitations are more clear, then field techniques can be systematically evaluated.

Photography is a long pursuit. Your best picture this month may not make the cut in two years...or two months...or two days. Taking a longer patient perspective is potentially helpful. Your best pictures are in the future.

\$\endgroup\$
3
  • \$\begingroup\$ I have used and still do use in-camera darkframe subtraction, sometimes. However, it just isn't feasible, often. Most tutorials for astro-scapes/astrophotography advise against it. It's not an option when you're stacking images, either, as the time between shots becomes to long, and stars wander too far. Moreover, it's problematic in higher ambient temperatures, especially with the K1 sensor (identical with the D800). With high ambient temperatures and long sessions, shooting darkframes for every shot = longer sensor runtimes = more noise. That's why I wanted to try manual dark frames. \$\endgroup\$
    – Kaschmir
    Apr 17, 2021 at 9:37
  • \$\begingroup\$ @Kaschmir Cameras have improved quite a bit in the last few years. In theory, the longer run time with in camera dark frame removal could generate more noise. Without testing, there’s no way of measuring if the difference is meaningful. Indeed, with the prevalence of video cameras have become so much more efficient that the seven minutes it took to make the images don’t generate enough noise to benefit from dark frame subtraction. At face value your question suggests this...or if dark frame subtraction makes pictures worse why do it? \$\endgroup\$ Apr 17, 2021 at 13:26
  • \$\begingroup\$ Why do it? Because I want to print as big as possible, and so far, I am not happy with the results. However, I haven't approached big enough stacks to reduce noise on that route, yet, so that might be the venue to go down. Regarding darkframes/in-camera darkframes for noise and hotpixel reduction etc, I followed discussions of other people on my sensor performance, and the verdict seems to be especially during summer, one of those measures has to be taken. \$\endgroup\$
    – Kaschmir
    Apr 17, 2021 at 15:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.