New script for denoising linear monochannel images

[mschuster]

Hi Joel,

1) You can crop before denoising. The flatfield must be cropped likewise. With DynamicCrop, set the crop bounds on the image, but before committing, drag the process icon to the flatfield, then commit on the image. Both should be cropped the same way.

2) Select and copy all of the lines in the Process Console window generated by ImageIntegration, use the Script Editor window to paste the lines into a new .txt file, save, and then load in the script. The denoiser will set Variance scale to account for the average frame normalization and weighting.

Thanks,
Mike

[joelshort]

Got it, thanks Mike.  Is there any reason, as far as MureDenoise is concerned, that I SHOULD crop before running the script?  Will black borders from registration negatively affect the denoising? 

[mschuster]

Hi Joel, this is something I have not tested. Guessing I think the script will work fine with black borders. Maybe you could try and let me know if you see anything bad? I will also test by adding a synthetic black border on my data.

Thanks,
Mike

[mschuster]

Hi Joel,

A quick test with synthetic black borders on one of my images: The script works fine with black borders.

I added a black border around my image, and a matching border on the flatfield, not black but rather a fill color of 0.5, as a zero in a flatfield would not be valid.

The denoised results look great. I generated the method noise image and see no artifacts in the image, along the border edges or elsewhere. In the black border areas there is a extremely small random looking texture in the range of 0.0 to ~0.01 DN visible only with an extreme stretch. This is of course negligible.

So the only restriction is no black borders in the flatfield. Denoising with black borders and no flatfield works fine likewise.

Thanks,
Mike

[joelshort]

Just to be clear, I understood you to say previously that one should use a flat if denoising a stacked image, correct?

I just can't believe what kind of results I'm getting with this script.  It's so easy and appears to work really well, as well as correctly handling signal vs. noise. 

[mschuster]

Thank you Joel,

Using a flatfield is optional. A flat can be used either on a single frame or on a stacked image. Either way specifying a flat helps the script account for differing noise levels due to vignetting. This is more useful on telescopes with lots of vignetting. When vignetting is small the improvement is negligible.

Regards,
Mike

[joelshort]

OK Mike, I've got a new one for you.  I'm currently working on an image of M42.  The main Red stack has 12-15min subs.  Of course I needed to take shorter exposures for the core, so I also took 12-1min and 12-30sec exposures for an HDRCombination.  In the .zip file I've linked to below there are 3 files:

"Orion_R" is the original stack with 12-15min images
"Orion_R_HDR" is the HDRCombination stack including 12-15min, 12-1min and 12-30sec images. 
And a master FLAT

When I used MureDenoise on the "Orion_R" image I get a very pleasing result.  Here were the parameters I used:
Gain = 0.335
Gaussian Noise = 16.02
Offset = 0
Variance Scale = 1.0
Cycle Spin Count = 16
I also used a flat.

Now here's the problem: when I use these same parameters on the "Orion_R_HDR" image it turns out horribly.  I first tried lowering the Variance Scale all the way down to 0.1 but it was still ugly.  Realizing that the HDR image now contained 36 images rather than 12, I changed the Combination Count to 36 and with a Variance of 0.1 the result was a lot better but I still saw some artifacts. 

How does using HDRCombination mess up the denoising process?  It seems to me like the only parts of the short exposure images that would be added to the main image would be the core of M42 and would not really affect the overall noise of the main image, yet it really messed up MureDenoise.

Here's a link to a zip file containing the images:
https://www.dropbox.com/s/mely703edrtkzgs/Orion.zip?dl=0

[mschuster]

Hi Joel,

Sorry for the trouble. The script will not denoise HDR combinations properly. In general the script does not work properly on linearly or nonlinearly processed images. Processing alters noise statistics in ways the script cannot handle.

Regards,
Mike

[joelshort]

I knew that, but I guess I didn't think that the HDRCombination script really "processed" the image at all...guess I was wrong. 

[pengsloth]

I used the script on a HDR image. However what I did was run the script on the long exposure image only and then combined the denoised long image with the normal short exposure. Because the short exposure is generally much better signal to noise it turned out great. Though I guess if your short image was also noisy you could denoise it too before combining.

[joelshort]

Yeah, that's what I did too.  Worked fine.

[jerryyyyy]

Hi,

Have been working away with the script and generally very happy but thought, Hey if this is good on a regular L image, what about a Super_L, where the RGB are added in.  I have some 28 L subs and an additional 16 each of RGB. 

I am still getting noise reduction, but it looks "pasty" unless I set the G noise very low....  When I run the ImageIntegration Proc and compare for example the G noise they are 2.09 versus 0.94 (to the -4) for the L versus Super L. 

Does this make any sense (sanity check please). 

Keep up the good work!

BTW, I have drizzle files on these data if you have any suggestions, guess that will not help. 

[mschuster]

Hi Jerry,

Sorry for the trouble, MureDenoise will not denoise multifilter combinations properly.

Thanks,
Mike

[joelshort]

Mike,
The more I use MureDenoise the more I really like it.  Fantastic!  The one suggestion I might make is that when I load the variance scale from a txt file created from the ImageIntegration process there is a pop up window that states the variance scale and the number of images used.  The variance scale automatically gets changed in the main window, but the combination count does not.  If you're forgetful like me, I might forget to change the combination count or worse yet can't remember how many images are in the stack.  Auto-populating that field would save me from myself.

[mmirot]

Mike,
The more I use MureDenoise the more I really like it.  Fantastic!  The one suggestion I might make is that when I load the variance scale from a txt file created from the ImageIntegration process there is a pop up window that states the variance scale and the number of images used.  The variance scale automatically gets changed in the main window, but the combination count does not.  If you're forgetful like me, I might forget to change the combination count or worse yet can't remember how many images are in the stack.  Auto-populating that field would save me from myself.

Can't we get the variance and number images put in the image header with XSIF after integration?
The script could automatically populate this info. Juan should jump at the chance for another use of XSIF

Also, can some post a screen shot on how to make the integration txt file?

And Mike this is a kill app!