New Version of BatchPreprocessing Script with Bayer Drizzle Support

[whwang]

1. Noise levels as reported by NoiseEvaluation script were 50-60% higher for the bayer drizzle R and B channels, and 25% for the G channel.
4. Upon several iterations of color saturation (Curve S), the bayer drizzle integration behaves a lot better, showing much less color noise visually than the standard integration (see attached image). I find this hard to reconcile with the numerical noise estimates. Juan?

Hi,

Let me comment on this one.  Unless something is fundamentally wrong in the data processing, images
(drizzle or not) made with the same amount of raw data should contain the same amount of information and noise.  The result that you see slightly lower noise level in the non-Bayer-drizzle version is caused by the interpolation, which acts like blurring the image a little bit.  In other words, the lower noise in the interpolated image is just an artifact or dilution.  You Bayer-drizzled image is as good as the other one.

Cheers,
Wei-Hao

[Astrocava]

Here is my attempt with 72 wide field shots. Not sure if I see any significant difference between Bayer Drizzle and traditional integration...
Georg

I don't see any difference too. Have you tried to substract one from the other?

[NGC7789]

I got a similar result with my widefield shots. Couldn't tell the difference. What is interesting though is that when I choose scale of 2 I couldn't tell the difference either except it is twice as big of course. Is drizzle more about fancy upscaling (i.e. larger output) rather than higher resolution (i.e. more detail revealed) or am I missing something because my data aren't sufficient to reveal the power of the tool?

[Ignacio]

1. Noise levels as reported by NoiseEvaluation script were 50-60% higher for the bayer drizzle R and B channels, and 25% for the G channel.
4. Upon several iterations of color saturation (Curve S), the bayer drizzle integration behaves a lot better, showing much less color noise visually than the standard integration (see attached image). I find this hard to reconcile with the numerical noise estimates. Juan?

Hi,

Let me comment on this one.  Unless something is fundamentally wrong in the data processing, images
(drizzle or not) made with the same amount of raw data should contain the same amount of information and noise.  The result that you see slightly lower noise level in the non-Bayer-drizzle version is caused by the interpolation, which acts like blurring the image a little bit.  In other words, the lower noise in the interpolated image is just an artifact or dilution.  You Bayer-drizzled image is as good as the other one.

Cheers,
Wei-Hao

I concur that debayer interpolation acts as a low-pass filter, and I was not surprised to see higher noise on the bayer-drizzle sample. What was a bit disconcerting was the slightly better FWHM, and the much worse color noise when saturated. This may have to do with specifics of VNG interpolation.

In any case, this is very good news to me, because what I was looking for most of all with this new approach, was to avoid the color bleeding I get in stars when doing the standard workflow. I still have to fully process the Omega Cen bayer-drizzle data, but I like what I saw so far.

Ignacio

[whwang]

I got a similar result with my widefield shots. Couldn't tell the difference. What is interesting though is that when I choose scale of 2 I couldn't tell the difference either except it is twice as big of course. Is drizzle more about fancy upscaling (i.e. larger output) rather than higher resolution (i.e. more detail revealed) or am I missing something because my data aren't sufficient to reveal the power of the tool?

I got a similar result with my widefield shots. Couldn't tell the difference. What is interesting though is that when I choose scale of 2 I couldn't tell the difference either except it is twice as big of course. Is drizzle more about fancy upscaling (i.e. larger output) rather than higher resolution (i.e. more detail revealed) or am I missing something because my data aren't sufficient to reveal the power of the tool?

Hi,

I don't have a PI example yet, as I have been too busy since the release of
the PI drizzle tools.  However, I hope my DSS example here can convince you
that Bayer drizzle is not just a fancy but useless thing.  It does help to bring out
the high resolution an image is supposed to have.
(original TIFF is here:http://www.asiaa.sinica.edu.tw/~whwang/misc/fig20-2.2.tif)

Of course, to achieve this, your image really has to be very sharp from the
beginning. 

Cheers,
Wei-Hao

[Ignacio]

I did this comparison of fully processed data from my recent OmegaCent session (22 dithered frames), using bayer drizzle 1x vs drizzle 2x integration (reduced back to original resolution at the end). Not very scientific as both processes are not exactly the same but similar (both done on a best effort basis).

Based on this I can say that deconvolution behaved better with the bayer drizzled data (easier to parametrize and less color bleeding in stars). The final result shows sharper and better colored stars, but less of them in the background. This last thing may be the result of the somewhat different processing of background noise. I speculate that the luminance PSF used during deconvolution is more consistent with the PSF of each individual color channel, in the bayer drizzle case.

FWIW, I also tried bayer drizzle on a different data set (30 dihtered frames), and got color pixelation in some parts of the frame (??).

Ignacio

[whwang]

Hi Ignacio,

From your examples, I see that the Bayer-drizzled image is better than the normal-drizzled image
in all aspects. (The difference in background faint stars should be some processing artifact.) This confirms my thinking that Bayer CFA images should NOT be drizzled.  Instead, Bayer-drizzle is the right way to go for Bayer CFA images.

Cheers,
Wei-Hao

[fulatoro]

Having issues getting DrizzleIntegration working. I ran the BatchProcessing script according the instructions, but when I run DrizzleIntegration and add the drz files, I get the following error:

* Parsing drizzle data file 22 of 22:
D:/GDRIVE/Astrophotography/Lights/M51-Baader/ppDrizzle2/registered/bayer/Whirlpool Galaxy (M51)_300sec_1x1__frame9_c_d_r_b.drz
*** Error: No location vector definition.
*** Error: No image could be integrated.
<* failed *>

[fulatoro]

Never mind...I forgot to enable "Generate Drizzle data" during integration.

Moussa

[rockyraccoon]

The script errors out, if no darks/bias/flats are selected.
I don't understand why any of those would be necessary to drizzle an image (I would expect warnings, but no errors)
 but the script just quits with the following messages in the console:

Code: [Select]

run --execute-mode=auto "/Applications/PixInsight.app/Contents/src/scripts/BatchPreprocessing/BatchPreprocessing.js"

Processing script file: /Applications/PixInsight.app/Contents/src/scripts/BatchPreprocessing/BatchPreprocessing.js


run --execute-mode=auto "/Applications/PixInsight.app/Contents/src/scripts/BatchPreprocessing/BatchPreprocessing.js"

Processing script file: /Applications/PixInsight.app/Contents/src/scripts/BatchPreprocessing/BatchPreprocessing.js

*********************************************************************
* Begin calibration of light frames
*********************************************************************

* Searching for a master dark frame with exposure time = 120s -- best match is a master dark frame of unknown exposure time.

*********************************************************************
* Begin calibration of light frames
*********************************************************************

* Searching for a master dark frame with exposure time = 120s -- best match is a master dark frame of unknown exposure time.
Interestingly the script claims best match for a dark frame that doesn't really exist.

[gvanhau]

Hello
I could finally test it.

The results are very good in my  modest opinión.
I attached a zoom 3:1 of a crop of a normal integrated image (left) and the same but Bayer drizzle integration on the right of a set of 24 pictures I took using a 24mm Lens on an canon 450D camera of the galactic center.
The stars in the Bayer drizzle image are rounder, showing an increase in resolution.
I dont see a increment in noise

Thanks Juan (and team) for this excelent tool.

Geert

[james7]

Am I correct in assuming that the only way you can currently use the Bayer Drizzle is with the BatchPreprocessing script? That is, there is no tool specifically to produce a set of files that have been reconstructed into RGB images using the Bayer Drizzle technique.

[Juan Conejero]

Am I correct in assuming that the only way you can currently use the Bayer Drizzle is with the BatchPreprocessing script? That is, there is no tool specifically to produce a set of files that have been reconstructed into RGB images using the Bayer Drizzle technique.

Yes, you're correct. We plan on releasing new tool versions to support Bayer drizzle outside BatchPreprocessing. They will be released as updates to the upcoming version 1.8.3 of PixInsight.

[NKV]

We plan on releasing new tool versions to support Bayer drizzle outside BatchPreprocessing. They will be released as updates to the upcoming version 1.8.3 of PixInsight.

Juan, please do it faster.
Because BatchPreprocessing don't provide full of power manual calibration and do it wrong. For example: BatchPreprocessing calibrate MasterFlat which already calibrated . Impossible to disable calibrate it.

[bitli]

I am afraid that this is an old known issue:
http://pixinsight.com/forum/index.php?topic=6936.msg47052#msg47052
But I never got an answer - the problem with BatchPreprocessing is that it is useless to use complex algorithms and 32 bit  images if we cannot be sure that the basic math is correct (or understand why it would do things differently).
-- bitli