PixInsight Forum (historical)

PixInsight => Release Information => Topic started by: Priami on 2013 November 05 03:59:24

Title: De-Blooming tool
Post by: Priami on 2013 November 05 03:59:24

Hi,
I am a new Commercial user. I would like to know if there is some tools (available in the current version or under development) to solve blooming issues.

Thanks!

Regards,
Leonardo Priami

Title: Re: De-Blooming tool
Post by: vicent_peris on 2013 November 05 07:41:26

Hi Leonardo,

You can solve the bloomings using shorter exposures and making a composition with the longer ones with the HDRComposition tool. I think Georg Viehoever a also wrote a tool to solve the same problem with gradient techniques, but in any case you're going to need the shorter exposures.


Regards,
Vicent.

Title: Re: De-Blooming tool
Post by: georg.viehoever on 2013 November 05 10:14:13

Leonardo,
have a look at thread http://pixinsight.com/forum/index.php?topic=5023.msg34770#msg34770.

In general, deblooming in the traditional way "paints" data into the spikes where simply not data exists, and therefore is not considered good practice. You can avoid them without "painting" by observational techniques, such as HDR exposures or rotated cameras.

Georg

Title: Re: De-Blooming tool
Post by: mschuster on 2013 November 06 10:19:44

Georg,

Just thinking about a deblooming integration with a rotated camera: Suppose sensor is square and frames are from two camera rotations, 0 and 90 degrees, all frames dithered.

Align all frames and create two integrations, one for each rotation. Compare integrations pixel-wise, marking bright pixels on one integration with dim counterparts on the other as "bloomed pixels".

Use these bloomed pixel maps as rejection maps in a new integration with all frames. In other words, integration does normal rejection as usual, but it also respects the bloomed maps as additional pixels to reject in the stacks.

The resulting SNR in the bloomed areas will be a factor of square root two less that elsewhere, assuming about equal number of frames in each orientation.

Run a noise reduction on the integration, using some sort of a merging of the bloomed pixel maps as a mask to slightly denoise only the bloomed areas.

Would this work?

Thanks,
Mike

Title: Re: De-Blooming tool
Post by: georg.viehoever on 2013 November 06 13:59:09

Your method would probably work. But I think simple integration with with rejection should also work if you have at least 3 rotation angles. In this case, 2 rotations will always overrule the bloomed pixels in the third one. I am not sure if it is possible to tune rejection in ImageIntegration such that it does the same with only 2 rotation angles in the source set.
Georg

Title: Re: De-Blooming tool
Post by: mschuster on 2013 November 06 14:30:36

I agree 3 rotations or more should work fine with the existing ImageIntegration, and likely only 2 is asking for trouble.

But more than 2 implies that larger areas have lower SNR, which would be nice to avoid.

Also another thought: maybe the existing II rejection map is useful as a denoise mask, that is, it identifies pixels that have lower SNR?

Thanks,
Mike

Title: Re: De-Blooming tool
Post by: georg.viehoever on 2013 November 06 15:16:31

If you do n rotations in n shots, the different SNR will be barely noticable (and significant only close to the star, where brightness is high anyway). Large n is probably unrealistic, since you also need n master flats, but n around 5 should be within reach....

Why dont you use HDRCombination (with a couple of short shots to fill in the bloomed areas) instead of rotated cameras? That also takes care of burned out stars. It probably has a similar SNR issue as 3 rotations, but if you dont have a rotator....

Georg

Title: Re: De-Blooming tool
Post by: mschuster on 2013 November 06 15:36:56

Worried that a couple of short shots will leave bloomed areas with poor SNR, especially when blooms cover dimmer nebulosity that require more time.
Mike

Title: Re: De-Blooming tool
Post by: pfile on 2013 November 06 17:07:37

Quote from: mschuster on 2013 November 06 14:30:36

Also another thought: maybe the existing II rejection map is useful as a denoise mask, that is, it identifies pixels that have lower SNR?

!! great idea. i think you'd have to crank the rejection parameters way down to get something that looks more like your image (for me a good rejection map is only rejecting obvious hot pixels, cosmic rays, airplanes and the like). i guess then maybe smooth the rejection image and use as mask during TGV or similar.

rob

Title: Re: De-Blooming tool
Post by: Priami on 2013 November 07 00:32:54

Thanks for reply to everyone.
One question: I have an Apogee Alta F9 (KAF6303), so this is NOT a square sensor. If I use rotation solution I will lose the 35% of area of the sensor. Or not?
Thanks
Leonardo

Title: Re: De-Blooming tool
Post by: georg.viehoever on 2013 November 07 01:27:38

Yes, you would loose some effective sensor area. You can reduce the loss by doing rotations<90 degree.
Georg

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 08 14:59:01

HDR does work to eliminate the blooms but it does present another problem.  The HDR images are 64 bit and when attempting to display them in the linear state they can be heavily posterized.  If one wants to use TGVDenoise, decovolution, et.al (which need to be done while the image is linear) it can be very difficult to optimize the parameters as it is difficult to see what was done through the posterization.  To optimize, one must iteratively apply the tool, perform a non-linear stretch, evaluate the results and repeat over and over again.  This can be very time consuming.

All of us in the user community anxiously await a release of Pixisight that provides for more than 16 bit resolution for the STF to significantly reduce the posterization of 64 bit images and hence improve our processing.

Cheers,

Don Barar

Title: Re: De-Blooming tool
Post by: pfile on 2013 November 08 15:14:41

hang in there :)

Title: Re: De-Blooming tool
Post by: vicent_peris on 2013 November 08 15:18:30

It's coming, now working fine in my computer.

V.

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 08 17:08:18

Hi Vicent,

Thanks for letting us in on the secret.  Might I ask when you think it will be ready to release?

Cheers,

Don

Title: Re: De-Blooming tool
Post by: vicent_peris on 2013 November 08 17:38:28

Well, it depends completely on Juan... I guess we'll have a new release in the next weeks. It will have 24-bit STF.


V.

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 08 18:40:22

24 bit STF!!!!!  That's even better than the 20 bit that Juan quoted a few threads back.  Yippee.

Don

Title: Re: De-Blooming tool
Post by: vicent_peris on 2013 November 09 06:35:24

My Iris nebula (HDR from 1200s to 0.7s exposures) is still somewhat posterized at 24 bits :D ).

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 09 07:22:19

Hi Vicent

What are the limiting factors that prevent developing an STF resolution higher than 24 bit?

Cheers,

Don

Title: Re: De-Blooming tool
Post by: vicent_peris on 2013 November 09 07:34:03

I think we don't need more bits than 24... The Iris is a very special case.

The limiting factor is the LUT table size and the time to calculate it. A 24-bit LUT table has a size of 16MB per color channel (this is independent from image size). A 28-bit LUT table would have a size of 256 MB per channel, so for each opened color image you would need an extra of 768 MB RAM space (even if the image is 1 mp size).

24 is a very good option IMO. It's a good compromise between visualization accuracy, LUT table size and calculation time.


Best regards,
Vicent.

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 09 08:11:48

Hi Vicent,

Thank you for the explanation.

I have studied your processing of the Iris Nebula (Dynamic Range and Local Contrast Tutorial), learned much, and have incorporated many of your concepts into my own image processing.

That said, what makes your Iris Nebula a special case that makes it more prone to posterization when applying the STF to the linear image.

Don Barar

Title: Re: De-Blooming tool
Post by: vicent_peris on 2013 November 09 10:37:10

Imagine the bloomings of a 7th magnitude star with a 1.23 meter telescope and a CCD camera with a QE of 90%...

The shorter exposures were about 0.7 seconds while the long ones were 20 minutes.


V.

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 09 10:57:30

Hi V,

Would I be correct in saying that the amount of posterization is related to the ratio of long/short exposures for the HDR composition?  In your example the exposure ratio is approximately 1700.  I know each subject is going to be different depending on what one is attempting to accomplish.  Based on your experience what would you say is maximum HDR exposure ratio that one can use and have minimum posterization?  Or, am I being really naive and attempting to oversimplify?

Don Barar

Title: Re: De-Blooming tool
Post by: Juan Conejero on 2013 November 14 11:59:21

Hi Don,

24-bit STFs are now working perfectly fine in the current development version of PixInsight 1.8.0. We are very close to the public release.

The specific case of NGC 7023 that Vicent has mentioned is a very extreme case. In this screenshot you can see how the image looks like with the traditional 16-bit STF (left) and the new 24-bit STF (right):

(http://forum-images.pixinsight.com/20131114/24-bit-stf-7023.jpg)

This image still shows a little bit of posterization in the darkest areas, even with 24-bit screen LUTs. To be completely free from posterization, this image would need a 28-bit STF, which is impractical for obvious reasons.

20-bit screen LUTs are sufficient for the vast majority of HDR linear images. In fact, we had implemented 20-bit STFs, which were working fine for all of our test HDR images, until Vicent uploaded this 7023 image for testing. This image "convinced" me of the necessity to reimplement STFs using 24-bit LUTs instead of 20-bit, but it is definitely a truly borderline case as I have said.

Title: Re: De-Blooming tool
Post by: dgbarar on 2013 November 14 15:01:33

Hi Juan,

Thank you for posting Vicent's Iris Nebula.  It was good to see the impact of 16 vs 24 bit STF resolution.  It really does make a significant difference.  I have several 64 bit HDR images I holding up processing because I am having difficulty optimizing noise reduction parameters. We all are eagerly waiting the release.

Cheers,

Don Barar