Mure Denoise Script -- When to Apply

[niteman1946]

Hi Group,

John Hayes posted a great image of M66 on the CN forum.  And he credited at least some of the results to applying the Mure Denoise (script).
I researched this tool and found it genuinely does a fine job of smoothing the image background without (in my opinion) ruining the target.

In my case I applied the script to the integrated luminance image after crop and DBE.  This is while still linear.
However, I think the documentation recommends applying either directly after integration, or at latest directly after crop.

I tried prior to DBE, but the results were extremely grainy with loss of detail.  But post DBE gave a great look and was more effective than TGV.

And the rest of the processing went as expected.

I'll accept that my interpretation of the docs was wrong.  But any thoughts on that?

Here's a quick screen print.  TGV processing on the left, MURE on the right.

Thanks,

Mark

[msmythers]

Mark

Here is a posting from Mike Schuster about when he suggest to use his Mure Denoise tool.
https://pixinsight.com/forum/index.php?topic=9206.msg59107#msg59107

I would suggest looking at the entire posting as there is much information about the tool and it's progression through versions. Searching for Mure in the forum will turn up a couple of pages worth of results also.


Mike

[Geoff]

The one on the left looks better. The one on the right looks to have too much smoothing.

[mschuster]

Yes, the script is designed to be used immediately after integration. Any further processing risks invalidating its results. The script uses a noise model that accounts reasonably well for the impacts of the calibration, alignment, and integration processes. Unfortunately it just can't deal with other modifications, even if the results remain linear. I suggest double checking script parameters. As a sanity check you might also try denoting a single calibrated frame. Don't expect the results to be "smooth". If that is what you want don't use the script.

[niteman1946]

Thanks Guys for your responses.
Not what I had expected, but that’s OK.

I’ll do more reading on this for the future.

Mark

[niteman1946]

Just a follow-up.

Here's a screen shot of M66 hot out of the Integrate oven.  No crop, no DBE on the left and only Mure Denoise on the right.
This looks good to me and behavior appears to be consistent with the documentation.

Notice the Mure Denoise process window shows Combination Count of "16".  The number represents the number of subs integrated and I took the Combination Count to require that value.  I had not picked up on that with my first post.

On a related note there is an odd behavior with PI on my desktop.  After the software has been active and worked fairly hard for an extended duration, the software gets hinky.  If the process that mis-behaves is something I'm familiar with, I know to shut down PI and maybe reboot.  In my original post I suspect that was occurring, and a reboot would likely have corrected the problem. 

Not blaming PI.

Mark

[mschuster]

Mark, please double check the script's Detector Gain and Gaussian noise settings, the values you used might be wrong. The script's documentation gives hints on how to determine these settings, using either camera / detector specifications or measurement scripts (for this you need to supply a couple bias frames and a couple flat frames).

Ultimately, the script is guessing as to what is noise and what is signal. Good guesses depend on good parameter settings.

[niteman1946]

Hi Mike,
thanks for your reply, and here are a couple of screenshots.

Image A:
Shows the original integrated image on the left with the Mure'd image on the right.
Note that FWIW, the Script window settings were as the script was first executed (i.e. no relation to my camera).
This should be exactly as my last post, and there is a slight smoothing of the background from the original to the Mure'd.

Image B:
Original on left and Mure'd on right.
This script features the Gain of 0.41 and Gauss of 17.07 as per my understanding of the Atik 383L+mono CCD properties.
Best I can tell there is no smoothing from the original to this Mure'd image.

So can you give just a little more guidance on what has been accomplished, if anything.

Thanks,
Mark

[niteman1946]

 Here's the Image B

[mschuster]

Hi Mark,

Ok, detector parameters seem better. If that is the method noise image in the background, the script did in fact do something as this image is nonzero. Method noise is the difference between the input and output. Method noise is useful as it provides the noise "guessed" by the script on a per pixel basis.

Note that blank sky typically has a bit of texture (eg. unresolved stars, etc), which is "signal" and which the script tries not to remove. But this requires that the script be set up properly and that all processing (calibration, alignment, integration) be "correct".

After selecting your input image in the script, click on the "Load variance scale..." button. If the script finds certain metadata in the input image, it will update the "Variance scale" value with a better estimate. If this metadata is missing, you will be asked to choose a log file, just cancel if this happens.

Beyond that, it is hard to for me to tell much just with your screen shots. If you want, and if possible, zip together 4 images: the integrated input image, one raw frame, its calibrated version, and its aligned version, and message me with a dropbox link. I may be able to find something else to suggest by looking these images.

Mike