Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Comparing a few different OSC processing techniques on M51

astrophotography dslr imaging
  • Please log in to reply
11 replies to this topic

#1 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 02:06 AM

I've processed my data of M51 taken with my Canon t3i DSLR a few different ways. Sharing the results of three different OSC processing techniques for comparison.  Particularly surprising was the results of LRGB processing (3 below). Feedback + insights welcome...

 

 

The same subs were used in all three images below (astrobin links in headings). They've have had identical processing steps applied with mild differences to parameters within processes tuned to the differences in the starting points:

 

1. Regular integration of debayered subs:

This is the standard OSC process of integrating debayered subs.

 

2. Bayer drizzle integration
Decided to try this since I had 62 subs: presumably statistically sufficient (with dithering) for individual R, G, B eval of pixels vs. interpolation using VNG debayering. I feel this results in an image with less noise and mild improved resolution.

 

3. LRGB processing
On a whim and having done no real reading on this, I decided to split my Bayer drizzled integrated image into L,R,G,B components using PI's ChannelExtraction. Then processed by pretending to have acquired these channels separately: combining RGB channels and then adding in L using PI's LRGBCombination.

I was pleasantly surprised by the LRGB processing result. I find it to be the most pleasing. It results in the deeper blues I have seen in most other M51 images, greater resolution and less overall noise. To be clear, I used the same Photometric color calibration settings and later applied the same color saturation on all three.
I'm curious about applying this more regularly to my other images. 

 

Thanks for looking.


  • Jim Waters, psandelle, elmiko and 2 others like this

#2 ChristopherBeere

ChristopherBeere

    Vostok 1

  • -----
  • Posts: 104
  • Joined: 24 Jul 2019
  • Loc: London UK

Posted 24 September 2019 - 03:44 AM

LRGB is the clear winner for me. Gonna give that a go myself.


  • arvindsh likes this

#3 Tapio

Tapio

    Vanguard

  • -----
  • Posts: 2135
  • Joined: 24 Sep 2006
  • Loc: Tampere, Finland

Posted 24 September 2019 - 03:50 AM

Color is more natural with LRGB but there's more detail in drizzled version.


  • arvindsh likes this

#4 whwang

whwang

    Mercury-Atlas

  • *****
  • Posts: 2837
  • Joined: 20 Mar 2013

Posted 24 September 2019 - 09:40 AM

Hi,

 

It's always good to experiment, and it will be even better to understand the principles behind the experiments.

 

Astrobin is slow at this moment, so I cannot actually see those images.  However, I want to point out that since all of them come from the same set of data, they should all eventually lead you to the same smoothness and color if your processing skills are good enough.  The only exception to this is that Bayer drizzle should usually (not always) give you sharper images comparing to non-drizzle-stacked images.  A drizzled image may appear more noisy, but this can completely be fixed in a later stage.

 

The LRGB trick you are doing here is not the real LRGB synthesis for mono sensors.  What you did here is essentially Lab processing for regular RGB images.  For example, PI's several noise reduction routines allow you to apply different strengths of NR to the luminance (L) and chrominance (ab) channels.  The allows you to preserve the sharpness and details in L while smoothing out chromatic noise dramatically.  This is not an LRGB workflow, but it works very well on any RGB OSC images.  Another example is that you can apply sharpening to L but not to ab, to increase the sharpness without amplifying the chromatic noise.

 

You don't need to pretend you are doing LRGB synthesis.  All you need is to realize some sharpening and smoothing are more suitable for luminance while others are more suitable to chrominance. Even if you apply the same kind of processing to L and ab, you can apply different strengths to them or with different masks etc.  Once you can master this, you will find it extremely powerful.  I just wouldn't call it LRGB processing.  Without real luminance data taken with a real L filter, it's not LRGB.  

 

My 2 cents.

 

Cheers,

Wei-Hao


Edited by whwang, 24 September 2019 - 09:42 AM.

  • arvindsh likes this

#5 Jim Waters

Jim Waters

    Mercury-Atlas

  • *****
  • Posts: 2982
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 24 September 2019 - 01:14 PM

I like image 3.  Real nice processing.  Can you provide details on the steps you followed.


  • arvindsh likes this

#6 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 04:02 PM

Astrobin is slow at this moment, so I cannot actually see those images.  

Thanks for the comments Wei-Hao. Astrobin was slow earlier but seems to have recovered. So hopefully you have a chance to see the images.

 

 

 

The LRGB trick you are doing here is not the real LRGB synthesis for mono sensors.  What you did here is essentially Lab processing for regular RGB images.  For example, PI's several noise reduction routines allow you to apply different strengths of NR to the luminance (L) and chrominance (ab) channels.  The allows you to preserve the sharpness and details in L while smoothing out chromatic noise dramatically.  This is not an LRGB workflow, but it works very well on any RGB OSC images.  Another example is that you can apply sharpening to L but not to ab, to increase the sharpness without amplifying the chromatic noise.

 

You don't need to pretend you are doing LRGB synthesis.  All you need is to realize some sharpening and smoothing are more suitable for luminance while others are more suitable to chrominance. Even if you apply the same kind of processing to L and ab, you can apply different strengths to them or with different masks etc.  Once you can master this, you will find it extremely powerful.  I just wouldn't call it LRGB processing.  Without real luminance data taken with a real L filter, it's not LRGB.  

 

Yes, I understand that quite obviously this is not LRGB processing and that careful use of other PI processes should give me the same results. Thank you for confirming that. My point with the 'LRGB trial' was that I was able to get to a better end state with this approach without having to fundamentally alter any of my other NR or sharpening steps. What is also not clear to me is why the colors would be different despite using the same color calibration process and colorsat settings. 



#7 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 04:15 PM

I like image 3.  Real nice processing.  Can you provide details on the steps you followed.

 Hi Jim,

These are the steps and associated PI processes used.

 

A. ChannelExtraction for RGBL components.

On RGB:

  1. RGB combination (after LinearFit)
  2. PhotoMetric ColorCalibration
  3. Deconvolution with external PSF
  4. Noise Reduction with MLT
  5. NL Stretch with HistogramTransformation 
  6. HDRMT on clone and then blending back to RGB with PixelMath
  7. ColorSaturation
  8. Contrast Boost with Curves
  9. NR with MLT
  10. Sharpening of galaxy with MLT

Perform a similar set of operations on the L image (minus color calibration and color saturation)

 

Blend L into RGB with LRGBCombination.

 

Hope this helps.



#8 sharkmelley

sharkmelley

    Vanguard

  • *****
  • Posts: 2325
  • Joined: 19 Feb 2013

Posted 24 September 2019 - 05:40 PM

An interesting experiment!  I like your systematic approach to processing, just like your systematic approach to understanding your sensor.

 

A few comments:

  • Why are there obvious colour differences between the versions?
  • The Drizzle version is definitely sharper than the Regular version
  • The "LRGB" version suffers from sharpening artifacts around the stars - dark halos

The LRGB version is the best of the 3 - it has the best colour balance, contrast and sharpness but the dark halos need dealing with.  It also has fine grained "natural looking" noise.  I dislike the plastic look or blotchiness of heavily reduced noise.

 

You should also try Bayer Drizzle with 2x scaling.  It's my standard processing workflow for 2 reasons:

  • The noise has a finer granularity which means it is easier to deal with. Do the noise reduction before 2x2 binning back to the original size
  • The 2x2 binned result is often slightly sharper than the Bayer Drizzle with no scaling

Mark



#9 whwang

whwang

    Mercury-Atlas

  • *****
  • Posts: 2837
  • Joined: 20 Mar 2013

Posted 24 September 2019 - 08:54 PM

What is also not clear to me is why the colors would be different despite using the same color calibration process and colorsat settings. 

 

It shouldn't be different, if the processes were properly implemented in the software.  Here I can think of two possibilities:

 

1. The normally stacked image and Bayer-drizzle-stacked image have different colors (balance between R, G, and B) to start with.  I noticed this behavior in PI, but this should be easily correctable in later stages. 

 

2. The "Photometric" color correction in PI (whatever it is, I never tried it) doesn't conduct photometry in an optimal way.  The apparently different (not intrinsically different) noise level in the drizzled and non-drizzled images may have some impact to the photometry and lead to different results.

 

You can see that at this level, the difference in the final result is not necessarily because some workflow is fundamentally better than the others.  It can be simply some flaws (or "features") in the implementation level of the processing program, or something else at the user level.  Later you may find something else that works better in the pseudo-LRGB mode, or the opposite.  It's hard to predict.
 

Well, I guess what I want to say is that if your find this "LRGB" workflow works for you, than use it, but don't stop there.  Keep exploring, learning new tools, and finding out why the other workflow doesn't work.  

 

Cheers,

Wei-Hao



#10 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 09:31 PM

An interesting experiment!  I like your systematic approach to processing, just like your systematic approach to understanding your sensor.

 

A few comments:

  • Why are there obvious colour differences between the versions?
  • The Drizzle version is definitely sharper than the Regular version
  • The "LRGB" version suffers from sharpening artifacts around the stars - dark halos

 

 

 

Thanks Mark—doing my best to build my understanding of how all of this works and share along the way :)

 

The color diff is baffling to me - esp. the Bayer drizzle vs. Bayer drizzle + RGB combination. My guess (similar to Wei-Hao's pt 2 above) is that PCC curve fitting is sensitive to some image parameters that are different between the two images I start with. 

 

The dark halos you astutely observe (mainly within M51) appeared when adding the L. The overall positive effects of adding L seemed to outweigh the presence of the halos. Open to suggestions on addressing.

 

 

 

You should also try Bayer Drizzle with 2x scaling.  It's my standard processing workflow for 2 reasons:

  • The noise has a finer granularity which means it is easier to deal with. Do the noise reduction before 2x2 binning back to the original size
  • The 2x2 binned result is often slightly sharper than the Bayer Drizzle with no scaling

Ah! I should've mentioned. In each case, I applied a 2X drizzle and after processing resampled to 50%—exactly as you suggest Mark, and for the same reasons. :)



#11 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 09:35 PM

It shouldn't be different, if the processes were properly implemented in the software.  Here I can think of two possibilities:

 

1. The normally stacked image and Bayer-drizzle-stacked image have different colors (balance between R, G, and B) to start with.  I noticed this behavior in PI, but this should be easily correctable in later stages. 

 

2. The "Photometric" color correction in PI (whatever it is, I never tried it) doesn't conduct photometry in an optimal way.  The apparently different (not intrinsically different) noise level in the drizzled and non-drizzled images may have some impact to the photometry and lead to different results.

 

You can see that at this level, the difference in the final result is not necessarily because some workflow is fundamentally better than the others.  It can be simply some flaws (or "features") in the implementation level of the processing program, or something else at the user level.  Later you may find something else that works better in the pseudo-LRGB mode, or the opposite.  It's hard to predict.
 

Well, I guess what I want to say is that if your find this "LRGB" workflow works for you, than use it, but don't stop there.  Keep exploring, learning new tools, and finding out why the other workflow doesn't work.  

 

Cheers,

Wei-Hao

 

Thanks Wei-Hao. I suspect PCC is sensitive to minor differences in initial image conditions. And point well made and well taken: this is all part of my efforts to better understand how this all works—so definitely not stopping at this. Also reading up on quite a bit of the math involved in image processing on the side too. I appreciate your inputs. Cheers!



#12 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 10:30 PM

You should also try Bayer Drizzle with 2x scaling.  It's my standard processing workflow for 2 reasons:

  • The noise has a finer granularity which means it is easier to deal with. Do the noise reduction before 2x2 binning back to the original size
  • The 2x2 binned result is often slightly sharper than the Bayer Drizzle with no scaling

 

Mark,

Forgot to mention, this thread and exchange with Juan on Bayer Drizzle made for fascinating reading and information. Thanks for all the hard work there in testing and helping fix it.




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics





Also tagged with one or more of these keywords: astrophotography, dslr, imaging



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics