Help: Can this noisy image be improved?

[cytan299]

Hi guys,
I'm wondering whether the image (see attached) can be improved. This image was processed in startools. The original image was taken with an OSC camera STF8300C with darks, bias and flats applied. Total integration time was (32frames*5 min) = 2 hours 40 min. As you can see, the result is still quite noisy. Is the solution to take more frames? Or is there some processing technique in Startools that will help with noise reduction. (I would prefer not to use Isolate because it always looks a little fake after using it. Well, unless there's a trick )

Any help and pointers are most welcome.

cytan

The original stacked frames after cropping plus Startools.log file are here:

https://www.dropbox.com/s/vwtec3ld6tyib1t/M101.zip?dl=0

Included in the M101.zip are:
cropped_M101.fit : the stacked file
startools_M101.tiff: The result after startools.
Startools.log: The log session

small_M101.jpg (384.75 KiB) Viewed 8438 times

[happy-kat]

Hi
What are you cropping it in, why not provide the original linear autosaved fits file instead please.

[cytan299]

Hi
What are you cropping it in, why not provide the original linear autosaved fits file instead please.

Not a problem. I used Nebulosity to Debayer with VNG. The stack was created using 10-90% tile because of stacking artifacts. There's still a few more that I cleaned up using Heal in Startools.

The stack before cropping in Nebulosity is called stackall.fit in the zip file. The Nebulosity cropped file is called cropped_M101.fit.

zip file:

https://www.dropbox.com/s/vwtec3ld6tyib1t/M101.zip?dl=0

cytan

[admin]

Hmmmm... That's disappointingly noisy for 2h40.
How bad is your light pollution? How does the stack look without darks/bias applied? How many darks, bias and flat frames did you take? Did you dither between frames?
It appears the darks aren't working very well, as I can see many hot pixels. Was the scope an achromat? Most stars appear purple once the image is color calibrated (they're throwing the Color module off initially, making the image way too green...).

The dataset is so noisy that AutoDev doesn't seem to be able to lock on to any detail very well. Only a manual Develop (I used the Home In feature) succeeds in suppressing the noisy background. Denoise reduction seems to be able to pin-point the noise quite well (as it should), and should be able to yield a smooth background;

stackall.jpg (74.26 KiB) Viewed 8401 times

When your data is noisy, it's best to keep your workflow simple;

--- Auto Develop
To see what we got.
--- Crop
Parameter [X1] set to [108 pixels]
Parameter [Y1] set to [26 pixels]
Parameter [X2] set to [3321 pixels (-69)]
Parameter [Y2] set to [2483 pixels (-52)]
Image size is 3213 x 2457
--- Bin
Parameter [Scale] set to [(scale/noise reduction 50.00%)/(400.00%)/(+2.00 bits)]
Image size is 1606 x 1228
--- Wipe
Vignetting preset (galaxy is small and nicely centered)
Parameter [Dark Anomaly Filter] set to [5 pixels]
--- Develop
Parameter [Digital Development] set to [94.59 %]
Parameter [Dark Anomaly Filter] set to [5.0 pixels]
--- Color
Parameter [Cap Green] set to [To Yellow]
Parameter [Dark Saturation] set to [4.30]
Parameter [Blue Bias Reduce] set to [1.19]
Parameter [Green Bias Reduce] set to [1.59]
--- Wavelet De-Noise
Parameter [Filter Type] set to [Distance Weighted Outlier Rejection]
Parameter [Grain Size] set to [12.2 pixels]
--- Wavelet De-Noise
Parameter [Scale 5] set to [50 %]
Parameter [Grain Size] set to [12.2 pixels]
Parameter [Smoothness] set to [80 %]

[cytan299]

Hi Ivo,
Thanks! Wow! A lot better! That's great processing and it's what I'm looking for .

To answer your questions:
I live in a very light polluted city, basically a red zone. Darks: 10, bias: 10 and flat: 10 frames. Dithering was used.

Perhaps I need more frames to get better results but I do think I need to redo my darks. The camera CCD might be deteriorating over time. I used bad pixel map to remove most of the hot pixels. There's about 0.2% hot pixels that were removed. And aligning and combining with 10-90%tile in Nebulosity which removed about 90% of the remaining hot pixels. What's left I used heal to remove in StarTools.

And I did use a light pollution filter in the image train. It's the IDAS LPS-D2 filter. I think that gives the colour problem.

My setup:
FSQ106 with IDAS LPS-D2 filter, STF8300C camera cooled to -15 degC.

From the dataset, I would think that doubling the dataset would be useful since noise would be reduced by 1/sqrt(2). Do you have other suggestions like:

• Stack method? I'm using average. Do you think Stdev1.5 or Stedev 2.0 will be better?

• a different way of aligning and combining? I'm using 10-90%tile

I'll try to reproduce your work this evening.

cytan

[happy-kat]

I find that creating a star mask (fat stars shrink grow grow can be effective) and to invert that so the stars aren't effected can help in the colour module if stars are otherwise turning purple/blue.

[cytan299]

I followed what Ivo had done but did not crop the image. Here's the results with the purplish stars:

startools_M101_ivo.jpg (312.13 KiB) Viewed 8386 times

[cytan299]

The same image (Ivo's settings) but recalibrated with eXcalibrator. The stars are no longer purple:

excalibrate_M101_ivo.jpg (320.8 KiB) Viewed 8385 times

[cytan299]

Used Ivo's bare settings, but doing all the enhancements in Startools: Contrast, HDR, Sharpen, Deconv, Life and then used eXcalibrator. Too fake?

excalibrate_M101_enhanced.jpg (296.69 KiB) Viewed 8385 times

[cytan299]

Hi Ivo,
Thanks a lot for looking at my data. It helped a lot!

Another question that I have is when do I use manual Develop rather than AutoDevelop? I've always used AutoDevelop and like you saw, it didn't work very well. So, is it better to try Develop first before using AutoDevelop and just try both and see?

cytan