Faint galaxy details in noisy images

[mstriebeck]

Despite our light pollution (downtown San Jose, CA), I keep trying to image galaxies. With lots (30 - 50 hours) if imaging time, I am getting some details. But when processing these, I seem to loose a lot of faint details. E.g. these two images:

M101:
HaRGB (after DBE, BackgroundNeutralization, ColorCorrection):

After processing:


M106:
RGB (after DBE, BackgroundNeutralization, ColorCorrection):

After processing:


My processing is usually the following:

• DBE (if necessary, multiple times)
• BackgroundNeutralization
• ColorCorrection
• TGVDenoise
• MaskedStretch
• Further Stretch with HistogramTransformation
• ACDNR (TGVDenoise usually leaves some blobs around)
• HDRMultiscaleTransform (to bring out details in the core)
• CurvesTransformation

If you want to inspect or play with the images:
M101: https://drive.google.com/file/d/0B0KcEqw-v71VbmJUVF9GQ2xabDQ/view?usp=sharing
M106: https://drive.google.com/file/d/0B0KcEqw-v71Vdzhtc1BOdzAyY1E/view?usp=sharing

(both are fit files after ColorCorrection)

Any ideas or pointers are welcome! Thanks!!!
     Mark

[msmythers]

Mark

Here is your M106. With more time I think it could be better or someone who knows what they are doing.


Mike

[mstriebeck]

Hi Mike,

Thanks - you definitively preserved much more detail then I could. Would you mind letting me know what steps you took?

Thanks
     Mark

[msmythers]

Mark,

Here are 2 icon sets. The first is the history for my image. I would not normally have used DBE where I did with your image. When I stretched your image the background was not flat. None of the masks are straight forward with defaults just to let you know.

The second icon is a set I use for smoothing the background. I use this sometimes because I shoot from a light polluted site also on the east coast of Florida, use very bad equipment and very short exposures. So my images are always noisy. The basic icon set came from someone in the forum here. I apologize for not remembering who. It's wonderful for what it does. I tweaked it for my more noisy images. I also added a convolution step. The main image needs to be named 'a' for the icons to work without change. Generally the range mask and the amount in the morphological tool has to be adjusted for best result.

The icons are listed in order 1-6. 1 and 2 are applied to the main image(a). 3 is applied to the created structure mask. 4 is applied to the main image(a) but the 'lower limit' and 'smoothness' will need adjusting for different images. 5 is applied to the created range mask. This range mask is applied to the main image(a). 6 is applied to the main image(a). The 'amount' might need to be adjusted. If my data was better quality and not mottled at times I would not need this icon set.

I don't think I've had 2 images that could use the same exact workflow ever.




Mike
http://www.astrobin.com/users/msmythers/
 

[pfile]

mark - i worked on your M101. the easiest thing for me to do would be to upload the project, but it is 1GB in size, so it will take a while.

i think that on both of these images the DBEs you had done, and the color correction were not effective. M106 is the worst out of the two. i think that you probably need to tune the parameters for both of these processes to get better results. for background neutralization, i generally make a preview over some part of the background, then tell BN to restrict it's work to that preview. using the readout i try to eyeball the RGB levels and set the threshold accordingly in BN. same with CC, but after BN the values tend to be a little different. what's crucial is to output the white and black reference masks and check that you've picked up enough of, and the right pixels. for DBE make sure that no sample is colored red and that each sample picks up enough pixels. it's usually worth moving the samples around to avoid any stars.

i kind of pushed M101 too far toward the blue here, but i tend to do things like that

rob

[topboxman]

It might be best to upload the stacked images without any kind of processing and let us do our own processing.

Peter

[mstriebeck]

I guess that makes sense (considering that I need to pay much more attention to DBE, BN and CC).

M101: https://drive.google.com/folderview?id=0B0KcEqw-v71Va0h1YkhyMm1PNHc&usp=sharing
M106: https://drive.google.com/folderview?id=0B0KcEqw-v71VUWg4eHV5ZkZwSlE&usp=sharing

Thanks!!!
     Mark

[mstriebeck]

Hi Mike,

I just tried your second icon set to smoothe the background. It worked pretty amazing! Much better then my attempts to smoothe the background with TGVDenoise. That's definitively a keeper!

Now, let me try to understand the first one...

Thanks
    Mark

[pfile]

i know it's a super pain in the butt, but i have managed skyglow gradients by running ABE on each sub (using ImageContainer) before registration and integration. i still had to run DBE a couple of times on the result, but getting rid of the gradients before integration really can be helpful, if only for the normalization to work better.

rob

[mstriebeck]

I tried that before (applying ABE to all subs) but I never noticed a difference.

[pfile]

well... it makes a difference. if there's a big LP gradient in the image then the scaling and normalization can be thrown off during ImageIntegration... but you have to try to tune the ABE carefully, and of course not every sub is going to be the same, so if you really want to get into it you may have to partition the subs into sets with similar gradients.

this image was acquired entirely from oakland 94618, and i had to employ the ABE technique or i don't think i would have been able to accomplish this result:

http://www.astrobin.com/68802/

rob

[topboxman]

Processing was challenging due to uneven gradients. Here is my quick and dirty process of M101 using RGB (no Ha):

Peter

[topboxman]

Here is my quick and dirty process of M101 with Ha added to RGB:

Peter

[topboxman]

I forgot to mention my processing procedure:

1) RGB Combination
2) Dynamic Crop of RGB and Ha
2a) Use NBRGBCombination script to add Ha to RGB for HaRGB processing.
3) ABE of both RGB and HaRGB. I tried DBE but I got better results with ABE.
4) Background Neutralization of RGB and HaRGB
5) Color Calibration of RGB and HaRGB
6) Multiscale Median Transform to reduce noise while images (RGB and HaRGB) are still linear
7) Histogram Transformation
8: Create Star Mask and use it for HDR Multi Scale Transform. I think I used 5 layers
9) Local Histogram Equalization. I believe I used Contrast Limit of 1.3
10) ACDNR to further remove noise for non-linear images
11) SCNR to remove any ugly green
12) Curves Transformation using "S" and then RGB/K
13) Tweak Histogram Transformation

Peter

[Warhen]

Mark, Are you defining a preview of dark sky for MaskedStretch background reference? And are you setting the target background high enough? The most apparent issue, and surprised no one commented- the processed images are clearly clipped as evidenced by an unnaturally dark background. Your dimmer detail is simply discarded. Perhaps this was intentional, but when compared with your great looking AutoStretch, you're losing a lot of data.  I much prefer HistogramTransformation for the initial stretch. Get a great STF AutoStretch visualization, then transfer that to HT for an excellent stretch. Good advice here by everyone!