DSLR NB Pre Processing Work Flow

[cdavid]

Hi everyone, I have read a variety of posts and other information on this issue with no conclusive opinion.  I have adopted this workflow for my Ha images obtained with a modified DSLR and Clipin Filter.  I am looking for some insights and general comments on this work flow.

1. Calibrate Ha raw files with a Master Bias, Master Dark, and Master Flat.

-one question that seems to come up regarding step 1:  Should the raw Ha files be calibrated as above in the usual fashion or should one use Split CFA and calibrate the Red channel from the Ha sub with only the Red channel from a bias, dark, and flat?

2. Debayer Calibrated Ha Subs with super pixel method and extract only the Red channels with channel extraction.

-will this provide the same result as using split CFA on the calibrated Ha Subs and then using CFA0?

3. Star Align and Integrate and proceed with processing.

Further, when one is shooting OIII subs in which both the blue and green pixels will have useful information:

Does one use split CFA to isolate the two greens and one blue channel and then combine them using channel combination and then proceed to align and integrate, or is it proper to just integrate them all together as separate blue and green channels?

Alternatively, instead of using split CFA can one debayer using super pixel then extract the blue and green channels and then proceed as above with either recombining the B and G together or just integrating them all as if they were the same?

Lastly, is there a tool that only extracts and discards lets say the Red channel and leaves you with a combined GB channel?

Thank you for looking!
Cheers
Carlos

[MikeOates]

Carlos,

Last year I posted a workflow for narrowband imaging with a
DSLR. SplitCFA was not arround then, but you may find it useful:

http://pixinsight.com/forum/index.php?topic=5748.msg39771#msg39771

Regarding:

Does one use split CFA to isolate the two greens and one blue channel and then combine them using channel combination and then proceed to align and integrate, or is it proper to just integrate them all together as separate blue and green channels?

Alternatively, instead of using split CFA can one debayer using super pixel then extract the blue and green channels and then proceed as above with either recombining the B and G together or just integrating them all as if they were the same?

Lastly, is there a tool that only extracts and discards lets say the Red channel and leaves you with a combined GB channel?

Keeping them separate will give you more options later as you may want to alter the blend between the two, or treat each separately for noise reduction.

Mike

[cdavid]

Thanks Mike, I was familiar with your previous post.  I've used a variation on it.

I just wonder if there is a true advantage to calibrating the light subs (Ha, OIII or Sii) with a Bias and Dark frame made up of the appropriate single channel?  Since we are dealing with just noise in the non-debayered image is it worth the trouble.  I see a clear advantage to doing this with the flat and in fact experimented calibrating an HA image with full channel flat compared to a stack calibrated with a flat made up only of the CFA0 (red pixel) channel and there was a clear and obvious decrease in the noise, in particular a peculiar diagonal banding noise I had encountered with the full channel flat.

I see your point about keeping them separate.  I just always think of OIII as a blend of blue and green signal that has been acquired in those green and blue pixels....I guess the SHO AIP script can be used to blend the green CFA and Blue CFA channels to create the OIII representation.

Carlos

[MikeOates]

Carlos,

Thanks Mike, I was familiar with your previous post.  I've used a variation on it.

Nice to know it helps.

I just wonder if there is a true advantage to calibrating the light subs (Ha, OIII or Sii) with a Bias and Dark frame made up of the appropriate single channel?  Since we are dealing with just noise in the non-debayered image is it worth the trouble.

Carlos

Well the individual colour pixels are in different columns, with the data flowing through different gates, so I would say it is worth while. Look at it this way, when imaging in Ha the red pixels are being used, so making a bias from all the other pixels will include the noise for pixels not being used. This could add extra noise rather than reduce it. The same will apply to the darks as well, regarding the noise.

And of course the size of the file using all the pixels in a bias will not match the single channel light anyway, so it won't work. And resizing the bias to match is just messing up the noise in the bias anyway making it completely invalid.

Mike

[cdavid]

Good points...it will work if you use all the channels as raw and extract the red channel after debayer ing. I think this makes perfect sense for example OIII which uses more than one channel but I guess I'm having trouble wrapping my head around an Ha exposure where we will toss the G and B channels anyways. If one uses the super pixel method does the noise in the other channels of the bias and dark really make a difference in the end since we will be tossing them anyways?

And to complicate further...I use a superbias instead of a master bias and hence at least with the bias the noise has been removed.

I may run an experiment using the various variations in the same set of data and try to measure the noise in the image and see what comes out.

Thanks!
Carlos

[xb39]

I may run an experiment using the various variations in the same set of data and try to measure the noise in the image and see what comes out.

Thanks!
Carlos

Hi Carlos,
did you run your experiment regarding different work flows.

At the moment I am dealing with the same questions as I have to process OIII narrowband images taken with my DSLR. My idea is as follows:
1.) split all raws with split cfa tool
2.) integrate the bias, flats, light for each channel (CFA1 - 3 separately; CFA0 will be discarded as the channel hopefully contains no information for OIII)
3.) a) combine both OIII channels (G and B) separately (after rescaling) with my RGB image using HaLRGB combine tool with prior channel-switch (see Geralds tutorial for this: http://www.werbeagentur.org/oldwexi/PixInsight/PixInsight.html ) CFA1+2 as green channel and CFA3 as blue channel OR
b) generate new OIII using mergeCFA-Tool and mix this image with my RGB

As I do not have an idea if this workflow will work all comments/ideas/hints/... are welcome! 

Best regards
Stefan

[cdavid]

Hi Stefan,

I have had such poor imaging skies the last couple months that I have not acquired enough OIII and SII data for the object I am currently working on.  I went ahead and processed what I had using two methods.

Object: NGC 7822

Data:
120 minutes HA using Astronomik 12nm HA filter with Canon 1000D (Astro-modified)
170 minutes SII using Astronomik 12nm SII filter with Canon T3i Full Spectrum
270 minutes OIII using Astronomik 12nm OIII filter with Canon T3i Full Spectrum.

My Ha data is quite noisy given that it was obtained using the Canon 1000D, my SII is very weak given the short exposure time.  I need to re-shoot the HA data with the T3i and obtain more SII data but despite all that I wanted to see what I had.

Method 1:

Calibrate Ha, SII, and OIII data with full RGB bias, Darks, and Flats obtained with appropriate filter.
Debayer Superpixel method, Extract R for Ha and SII exposures and extract G and B for OIII exposures.
Star Align:  For Ha and SII I used only the extracted R channels.  For the OIII data I used the G and B together as if they were monochrome exposures.
Integrate R(Ha) data for Master HA, Integrate R (SII) data for Master SII, and Integrate G and B data together for Master OIII

Combined the 3 channels using the Multichannel synthesis tool to create final image in HOO palette but with modification of using 20% SII in the R Channel and 20% Ha in the B channel.

Method 2:
Split CFA Bias, Darks and Flats, and lights.
For Ha and SII:  Calibrate, star align and integrate using CFA0 only.
For OIII:  Calibrate CFA 1, 2, and 3 with their appropriate CFA calibration frames. Then star align and integrate the CFA 1, 2, and 3 together as mono images to create a Master OIII.
Create Final Image as above using modified HOO palette.

What I found after measuring noise was that the difference between methods was negligible for the OIII and SII data which were fairly clean images.  Visually I could not tell them apart.  Using the Measure noise script, the second method had a very minimal improvement.

For the Ha data which was obtained with a much noisier camera the difference was dramatic.  The final result is still quite noisy but a peculiar diagonal banding noise was much better suppressed using method 2.  Here is an image comparing the two.

I did not think to use the combine CFA and may try that as well.  I also have not used the method in the video link you provided.  I need to give that a whirl as well.

In any event the final method 2 image is at:
http://astrob.in/144013/0/

Hope this helps.

Cheers
Carlos

[xb39]

Hi Carlos,
thank you very much for your detailed answer and your experiments. Method 1 would things make much easier as I only have to do the workflow once and not for each CFA separately (in case of the OIII channel for CFA 1-3).
BTW your NGC7822 looks great!

Best regards
Stefan

[xb39]

Hi Carlos,
here is the result of Method 2.

Cirrus Nebula in Bicolor (Ha + OIII):


Ha:


OIII:


More information and better resolution pics at astrobin:


Best regards
Stefan

P.S.: ...>120x600s RGB data are waiting for processing...

[cdavid]

Excellent!  Very nice detail.  So for confirmation are you finding much difference between he methods?
Thanks
Carlos

[gvanhau]

Hello Carlos and Stefan.

When you use Super Pixel Debayer or Split CFA to separate the red channel (or blue and green), you are loosing spatial resolution (unless you are already oversampled).
One way to go, in order to not to loose resolution, is to use bayer drizle integration and separate the R, G or B channel after the integration. Of course, for that, you need to have more light frames and they must be dithered, but I think that the results will be better.

Regards
Geert

[xb39]

Excellent!  Very nice detail.  So for confirmation are you finding much difference between he methods?
Thanks
Carlos

Hi Carlos,
Thanks. 
No, I din´t cross check with method 1.

Hello Carlos and Stefan.

When you use Super Pixel Debayer or Split CFA to separate the red channel (or blue and green), you are loosing spatial resolution (unless you are already oversampled).
One way to go, in order to not to loose resolution, is to use bayer drizle integration and separate the R, G or B channel after the integration. Of course, for that, you need to have more light frames and they must be dithered, but I think that the results will be better.

Regards
Geert

Hi Geert,
very interesting.
I´ve drizzeled the images after integration of the splitted CFA data.
I thought a major advantage of the above mentioned workflow is that the splited images were integrated separately which decreases or does not incorporate the noise of the other CFA channels (e.g. for Ha use only CFA0 data and discard CFA1 - 3).

Best regards
Stefan

[cdavid]

Hi Geert, Yes exactly as Stefan stated..... I also will drizzle some images.  The utility of the split CFA is in the processing of narrowband data where you are only employing the Red pixels for  Ha or SII and only the G and B for OIII.  You are already losing spatial resolution.  By using the split CFA we are eliminating the unwanted noise signal in the other pixels.  One can always upsample the image later especially if combining with RGB data.

Regards
Carlos

[gvanhau]

Hello

The reason why I recommend you to use bayer drizzle, is that with this algorithm/technique you can fill the missing pixels with real data (product of rotation and displacement of the other pixels under the same bayer color filter) , this can not be done with only upsampling the image.
Also, I dont think that the calibration process in osc images (DSLR) or al least should not propagate the noise from one color to another: it sould be the same calibrating first and then separate color as separating color first and then calibrate.

Regards
Geert