Back to forums

PixInsight Noise After Dynamic Background Extraction

Pleiades Astrophoto PixInsight Processing Background Extraction Beginner Deep Sky
5 replies•616 views
camxwest avatar
camxwest avatar
Hi All,

I'm brand new to PixInsight and are processing my first image in it. Totally bumbling my way around at the moment, but are excited by the potential once I learn how to harness it.

After I run a Dynamic Background Extraction, I seem to get a lot of background noise and wanted to check in to see if this is to be expected and I just need to try and reduce it in further steps or whether I can improve the outcome from the DBE. Below is an image taken directly from Deep Sky Stacker with DBE run on it only.

https://astrob.in/fvzsxo/0/

I'm in Bortle 6, M42 is fairly low and I'm shooting about 20 degrees over the top of streetlights with a ZWO294 and no filter, so there is a reasonable amount of light pollution to remove. Is this the kind of result I should expect straight after background removal?

Thanks in advance!
Respectful
Konstantin Firsov avatar
Konstantin Firsov avatar
Hi camxwest,
Looks like you are are facing an issue that has a fundamental background. It is caused by a Poisson noise, that can be observed in any discrete stochastic process, like rain or photon flow. The pixel value of background light pollution has it's mean value and a Poisson random component (noise). This noise equals a square root of the mean value.
When you remove background gradient with PixInsight or any other way, you remove the mean value, while the noise component still is there. You have to apply noise reduction or look for the darker skies smile
By the way, ASI294 is known for it's relatively high amp glow. You should probably have seen it in your master-dark. When you subtract the master-dark from your light image you remove only the constant amp-glow component, while the Poisson noise component stays in place. You can clearly see this amp glow noise in the calibrated light image if you stretch the histogram strong enough.

CS, Konstantin
Helpful Insightful
dkamen avatar
dkamen avatar
Hi,

DBE creates a relatively smooth background image and subtracts it from your original. Locally, this is very similar to subtracting a constant. Therefore it cannot (and should not) alter your image's noise profile significantly. What it does is make the noise stand out more because it remains the same even though the total intensity is reduced.

Some dummy figures for illustrative purposes:
Let's say your image has linear signal ranging from 40% to 50% ( a completely dark pixel is 0%, a completely white pixel is 100%). 30% of that is background and another 2% is noise. So noise is 2/40th to 2/50th of what you see in the preview.
When you subtract the background, signal is now ranging from 10% to 20% but noise stays the same. it is now 1/10th to 1/20th of what you see in the preview (2.5 to 4 times more distinctive).

Stretching further amplifies the "problem". The brightest parts of your original image were at 50% so they are multiplied by 2 to become 100%. After you have subtracted the background, the brightest parts are 20% so they are multiplied by 5 and so is their noise. Similar for the darkest parts. Note: I am seriously oversimplifying here for various reasons
  • , but the main principle holds true: stretching will tend to amplify noise and will do so more after you have subtracted the background simply because noise is a greater part of the signal now. Especially the "naive" stretching done by ScreenTransferFunction, which is designed to reveal as much detail as possible in the stretched image.

    • So basically what you see is perfectly normal and yes, this noise can and should be treated at later stages.

    [*] stretching is non-linear and affects different images differently and of course you can control it to a great degree: use a different stretching algorithm that results in a smoother image such as MaskedStretch, or just tune HistogramTransformation to clip some shadows and their associated noise etc.
    Well Written Helpful Insightful Engaging
    camxwest avatar
    camxwest avatar
    Konstantin and dkamen, thanks for your detailed responses. It's very much appreciated!

    I know where to focus my efforts now to see how much I can clean this up.

    Cheers!
    Well Written Respectful
    Leonardo Landi avatar
    Leonardo Landi avatar
    I've had a similar problem with my M33. In that case my backgruoun was compromised by a totally wrong-made flat field, that introduced strong gradients (even worse than LP in my opinion).
    I managed it with a first denoise run with John Rista NR method (have a look at it, it's super simple and you can perform it quite well even with the EZ Suite). After the stretch, I performed 2 hands of ACDNR. Then I masked the galaxy and used TGV again to remove the chromatic noise. The result is quite smooth, you can see it in my gallery. Hope this can help you.
    Helpful
    Elmiko avatar
    Elmiko avatar
    Another tool to reduce the background noise is MSLT using the built-in Mask. I use it after DBE and Color Calibration, and right before I stretch to Non-Linear. I think a lot of what you are seeing in your image of M42 is extended nebulosity.  Check out Shawn Nielsen Utube channel Visible Dark ca. He shows how to use MSLT.      Mike