Mosaic Assembly Steps

[shurik]

I would like to re-visit and re-post the subject that has been asked more than a year ago by another User but generated zero replies, maybe someone can help and answer these questions or provide directions to tutorials/processing examples how to properly handle mosaic panels and what is the sequence of processing steps?

Original post:

As you might have read in another threat here, I'm working on a mosaic of 6 full frame images. (Carina Nebula to Running Chicken)
I succeeded in aligning and merging the panels using the MosaicByCoordinates script. (thx for the tip on that one!)
Now I want to do a proper mosaic, but run into the challenge that the intensity AND color balance seems quite different between panels.
I've tried using the dnaLinearFit 0.1 found through the tutorial on trapped photons (http://trappedphotons.com/blog/?p=994)
but I get very 'strange' results and some clipped pixels. Strange meaning very strong color differences over the panels, with mainly a big difference between carina and the other panels.

So I thought I'd better start over and make sure the panels are proper before aligning them. However, I wonder what the best approach is here;
- Should I apply individual BackgroundNeutralizations or could I use a background preview on 1 panel to apply it on other panels?
- Should I apply Colorcalibration already? And then same question as before; each panel individually or use the preview of 1 panel on the other panels?
- the Carina panel has 260 minutes integration time, Running Chicken 160 and the other panels 90 minutes. Which panel is best to use as reference for any form of LinearFit? The brightest one or another?

I've not come to the stage of removing bright stars at the edges etc. , but have read up on all those tips. Just want to do this step by step, Step 1 being the alignment and ability to stitch the frames together .


Your help is very much appreciated!

[John_Gill]

Hi,

I am also busy with a large mosaic and faced a similar problem.  I receive a great tip from "MSmythers" on this site, about using "AutoHistogram" on each individual image before doing a "StarAlignment" and the "GradientMergeMosaic"  This seemed to even out the background.  With the SA, I used 2D-Surface Splines and removed the FrameAdaptation.  I also reduced the GMM feathering between images as that sometimes caused the join to blur too much.

I guess you will need to test the FrameAdaptation and Feathering to achieve the best result.

Look up
John

[shurik]

Thank you for those refrences!

 my general confusion is that if we assemble frames shot on different nights and during different moon phases, do we deal with those issues for each panel separately when images are still in linear form (non-stretched, no DBE or other normalization actions, no structure enhancement), or do we create one large mosaic image and then start the correction steps? I understand in Photoshop I would have used layers that way  I keep the mosaic and adjust each panel separately while its a part of the mosaic so I can visualize how it "fits" with regard to the overall image, but in PI there are no layers, so if I assemble mosaic first , how do I deal with individual adjustments for each panel?

[dnault42]

It has unfortunately been some time since I wrote that tutorial and things have changed a bit since then.  The Mosaic by coordinates makes aligning panels much easier and should be included by default.  The dnaLinearFit script is not as robust as the LinearFit process however it does handle retaining black pixels as black making it easier for GMM to know what is image data and what isn't.  If you get 'strange' results from it you may try some of the other fitting methods.  The default 'mean / median' works for a lot of cases, but not all, so you might try 'Least Squares' and 'Scalar' as well.

Generally when I build a mosaic with multiple filters I do one whole filter first.  I do whatever I have to do to get the intensities to match and to make sure the transitions are smooth.  Then I use that panel to register and linear fit the remaining panels for the other filters.  Doing it this way, you can use the 'Frame adaptation' feature of StarAlignment to match the intensities.  Once I've built all the mosaics for each filter then I treat it as one big image doing color calibration, etc. as usual.

I only resort to stretching the data prior to combining with GMM if there are stars or gradients that simply can't be solved correctly by cropping stars and doing DBE, etc. on the individual panels.  These are nonlinear stretches making it difficult to match things up after.

Regards,
David

[shurik]

thank you, do you know of any tutorials, or step-by-step processing examples for doing mosaics, starting from stacked but otherwise unprocessed panels?

[dnault42]

I believe Kayron has one that is a bit more recent than mine on his website (www.lightvortexastronomy.com).  Warren Keller also has some tutorials at IP4AP.com and has a whole chapter on mosaics in his new book (Inside PixInsight).  I think both of them use my dnaLinearFit script though.  There are almost certainly other tutorials as well, but I can't think of them off-hand.

As far as I know there is nothing inherently wrong with my tutorial other than some of the registration steps could be updated, but some of the other tutorials do go into more details.

Also, if you would like some help getting a good fit with some of the panels you could upload some of them for me to take a look at.

Regards,
David

[Juan Conejero]

The Mosaic by coordinates makes aligning panels much easier

It must be emphasized that the MosaicByCoordinates script, along with ImageSolver, is the best way to build large mosaics consisting of more than 3-4 frames. These wonderful scripts have been created and are being maintained by Spanish software developer Andrés del Pozo.

StarAlignment can create small mosaics of 2-4 frames, but to build a geometrically correct large mosaic, you definitely need astrometric solutions. MosaicByCoordinates is simple and efficient. You can also use StarAlignment to register all frames against a background reference image generated with CatalogStarGenerator, which also gives excellent results. An upcoming version of StarAlignment will implement a much more accurate and powerful local distortion correction algorithm, among other important improvements. I am currently working on this.

Unfortunately, I am alone with PixInsight core development and maintenance, so currently I have no time and resources (and stamina, I'm getting older!) to start and/or help new development projects. Otherwise I would like to integrate Andres' astrometric tools with the PixInsight Core application, including C++ implementations of his scripts.

The dnaLinearFit script is not as robust as the LinearFit process however it does handle retaining black pixels as black

That's great! We can implement this functionality in the LinearFit tool if you want. The ColorCalibration module, which LinearFit belongs to, is open-source and hence available at our GitHub repositories:

https://github.com/PixInsight/PCL/tree/master/src/modules/processes/ColorCalibration