Newbie processing techniques feedback

[manojkoushik]

Hello all, I am currently evaluating PI in the hopes that I can get better results than PS that I have been using till now. And so far it is very promising. With PS I usually only do some stretching with maybe an artificial flat via guassian blur to remove vignetting etc.

I tried PI on two stacks (all files at:https://www.dropbox.com/sh/9vvf00mls5bxr3n/4K4YVc7yYa , Both stacks done in DSS using auto adaptive weighted average, RGB channels background calibration and alignment of RGB channels in the final image)
- On Crescent I did the following: STF to aid in cropping, HistogramTransform to bring out details, HDRWaveletTransform, ACDNR, AutoHistogram
- On Cocoon I did the following: STF to aid in cropping, HistogramTransform to bring out details, ATrousWaveletTransform with noise reduce, AutoHistogram

Would like to hear from you if the workflow above looks good and if the final result looks good. I am sure you can achieve a better result from the same raw stack, I would like to see the processing result and also your workflow.

thanks,
Manoj

[manojkoushik]

I should add that for both stacks the crop was followed by ABE.

[pfile]

well i think you've done quite well - the original data is pretty rough. lots of noise and star trailing. the end results are a bit soft but with all the noise it's understandable.

one thing that's strange is - why are the stars looking like donuts?

you might get more out of this data if you calibrate, register and integrate with PI but working with less-than-stellar data is always hard (though it builds character!)

rob

[manojkoushik]

yeah very noisy data Shooting from the suburbs of San Francisco. The camera is OSC (QHY8) with a Nebular filter (DGM NPB). So luminance sensitivity is low. Combine that with a mount with about 30" of PE (corrected down to about 6" with PEC), and a guidescope setup (ES 127ED with 50mm guidescope), there is some drift because of the flex + tracking.

I believe the stars ended up that way because of I couldn't figure out the right deringing settings. That combined with the slightly oblong stars ended up causing that weird donut shape.

But otherwise I am on the right track?

Also is there any easy way to stack with PI when the images are from multiple nights (with it's own set of flats and darks)? Any tutorial on how do to this?

[manojkoushik]

Actually I notice the donuts on the fainter stars even in the stack. Could be something to do with the way I am stacking? I am using Drizzle and auto adaptive weighted average. I should probably try PI for the stacking itself.

[astropixel]

PI may produce better results stacking using the Batch Preprocessing Script.

[papaf]

Try removing the hot/cold pixel correction in DSS. My friend had the same problem.

[pfile]

yeah i am in oakland and unless i'm doing narrowband i have to do 8-12h of integration time per channel to get good results.

that kind of donut thing could be caused by pixel rejection, i suppose, if the stars are overexposed.

i guess if you are going to be stuck with such noisy images, you should take a look at TGVDenoise.

rob

[manojkoushik]

Tried PI for stacking and am really happy with what came out. I am not even sure if DSS was applying the flats correctly. Some of the artifacts which I was noticing after DSS stack are now gone. Could very well be cockpit error, of course.

New images are at https://www.dropbox.com/sh/k8htieai1biesjo/V-FWYYvuU_

Used Deconv to make things a bit tighter, ACDNR and ATrousWavelet to reduce noise and enhance some of the structures.

Rob: You are right, I have to capture a lot of data to get anything meaningful. The Cocoon is about 4 hours (2 hours of 10 min subs and 2 hours of 15 min subs). Planning to go narrowband soon. The Crescent is almost 5 hours. Have to restack that as well.

Would like to hear thoughts on two things:
- Given the latest am I on the right track in terms of workflow? Stacking is a bit more work with PI (since, if you have multiple nights of images, each night would have to be calibrated first before the register and stack, unless there is a way to do this with batchprocessing script that I don't know of)

- My camera is a QHY8 OSC. The bayer pattern I used was GBRG (with bottom-up) and noticed a red hue all over the image. This went away after stacking and ABE, and the colors in the final look correct. But I am still not completely sure if I am debayering correct. Everything else looks way too wrong (all other bayer patterns with up-bottom or bottom-up). Anyone with some input on this would be great.

[Juan Conejero]

Stacking is a bit more work with PI (since, if you have multiple nights of images, each night would have to be calibrated first before the register and stack, unless there is a way to do this with batchprocessing script that I don't know of)

What makes you think that? As long as you don't mix different objects and have proper bias, dark and flat frames, you can calibrate and align everything in a single instance of the BatchPreprocessing script. The script will select the required files automatically. Only if you need different flats for each night things can get a bit more involved. In this case it is often easier to separate calibration (but not alignment) in several batches. You can and *should* also integrate everything in a single operation because this way you'll get the maximum possible SNR in the final image. Astrophotography is all about pushing signal up and noise down.

New images are at https://www.dropbox.com/sh/k8htieai1biesjo/V-FWYYvuU_

In the second (stretched) image, the background has a strong magenta cast (look at the histogram and you'll see the green and blue peaks more or less aligned to the right of the red peak) and the rest of the image is too red. Both problems can be fixed very easily on the linear image (the first one, before stretching) with the BackgroundNeutralization and ColorCalibration tools.

Used Deconv to make things a bit tighter

I have two main advices:

- In case you haven't done so already, watch all videos on Harry Page's Astro Shed site: http://www.harrysastroshed.com/pixinsighthome.html

- Try to keep it as simple as possible. For example, deconvolution is a very delicate process requiring high SNR data to work properly. You don't need deconvolution for now. Try to apply a minimum of processing to achieve: (1) a flat illumination profile with DBE, (2) a correct color balance, which means a neutral background and a good white balance, (3) some judicious noise reduction in the linear phase with the MultiscaleMedianTransform and TGVDenoise tools, (4) nonlinear stretch with HistogramTransformation, (5) optionally, some noise reduction in the nonlinear stage with TGVDN and/or ACDR, and (6) optionally, some color saturation adjustments with CurvesTransformation. This would be a good starting point.

[manojkoushik]

What makes you think that? As long as you don't mix different objects and have proper bias, dark and flat frames, you can calibrate and align everything in a single instance of the BatchPreprocessing script. The script will select the required files automatically. Only if you need different flats for each night things can get a bit more involved. In this case it is often easier to separate calibration (but not alignment) in several batches. You can and *should* also integrate everything in a single operation because this way you'll get the maximum possible SNR in the final image. Astrophotography is all about pushing signal up and noise down.

I have separate flats for the various nights and so had to do calibration in several steps before a single registration and integration like you say. Having separate file groups in the script even for the calibration would make this easier. Other than that though it's great!

In the second (stretched) image, the background has a strong magenta cast (look at the histogram and you'll see the green and blue peaks more or less aligned to the right of the red peak) and the rest of the image is too red. Both problems can be fixed very easily on the linear image (the first one, before stretching) with the BackgroundNeutralization and ColorCalibration tools.

Does this mean the debayer settings I am using are correct? I was not sure about that. With DSS I have never had to play with the matrix settings so was not sure what exactly it is. Everything I read online says it is GBRG for QHY8 and I use Nebulosity 3 or Sequence Generator Pro to capture so they should not be doing anything funky.

I  have two main advices:

- In case you haven't done so already, watch all videos on Harry Page's Astro Shed site: http://www.harrysastroshed.com/pixinsighthome.html

- Try to keep it as simple as possible. For example, deconvolution is a very delicate process requiring high SNR data to work properly. You don't need deconvolution for now. Try to apply a minimum of processing to achieve: (1) a flat illumination profile with DBE, (2) a correct color balance, which means a neutral background and a good white balance, (3) some judicious noise reduction in the linear phase with the MultiscaleMedianTransform and TGVDenoise tools, (4) nonlinear stretch with HistogramTransformation, (5) optionally, some noise reduction in the nonlinear stage with TGVDN and/or ACDR, and (6) optionally, some color saturation adjustments with CurvesTransformation. This would be a good starting point.

Thanks Juan, I will try this out after I hear back from you guys that the debayer is correct.

[manojkoushik]

New stacks and processed images at https://www.dropbox.com/sh/s6mcc7kdh285cii/37sHfTN5bv based on feedback from Juan. Juan, would love to hear back from you on what you think. Stack and processing all done in PI. Basically followed the workflow you suggested.