Help Requested w/Gradient Merge Mosaic Problems

[jeffweiss9]

Georg-
  Just tried my first mosaic (2x2) following Steve Allan's video and my results could be a carbon copy of those of kerrywaz1 (except for a different target).  I have the same border star blow-ups and significant seams after optimizing the GMM parameters as much as I could. Did you ever incorporate your alternate method in a version that you might release?
Thanks,
-Jeff

[georg.viehoever]

No new version, sorry. Maybe you could provide some of your files so i can reproduce and maybe solve the problem...
Georg

[jeffweiss9]

Georg-
Thanks very much for the offer.
If you email me (address in my profile) your email address so I can link it to my Dropbox, you can see the input components there. I put two complete sets: one for the luminosity, that worked the best, and the other for the Blue filter, which worked the worst.  Pick whatever pieces you want to try. They all appear to have the same problem discussed in detail above with different severity - bright stars near the edge creating black and white comets and driving the gray levels way off.   I was able to sort of clean up the Luminosity mosaic (not really) but could do nothing with the RGB (I also have Ha I haven't even processed yet).
I attach here the GMM outputs for Lum and Blue, run with "best" 40 feather and 1 radius (max radius 10 was the same) but all settings gave similar results, a bit better or worse.  A1 is in the upper right and they run clockwise from that through a sliver added as A5 at the uppermost left.
Thanks very much.
Clear skies,
Jeff

[georg.viehoever]

Jeff,

thanks for providing the images. So far I have had a look at the luminance files (those with an L in the name), and I did not have a lot of trouble getting decent reults. Let me outline the steps:

1. Create reference mosaic: This mosaic is later being used to aligned the images for GMM. Use StarAlignment with Registration Model "2D Surface Splines" and Distortion Correction enabled (to adjust for slight distortions between the images). Set mode to "Register/Union Mosaic" and Reference View to one of the images, so that StarAlignment will produce a merge between reference image and the image you apply it to. Activate Frame Adaption so images get similar brightness. Apply this to one of the neighbors of the Reference View, make the result the reference view and repeat with the next image until to have a combination of all images. The original image and StarAlignment settings can be seen in screenshot 1, result in screenshot 2 (next post) left. The result is not bad, but we clearly see the borders between images.

2. Align images to reference mosaic: Use the same settings in StarAlignment as in step 1, but in Register/Match Images mode. Settings in screenshot 2 right. Result in images of screenshot 3 (next post). Tje aligned images show the parts of the image in the correct position

3. First GMM attempt with aligned images: Use the default GMM settings, see screenshot 3. The result is in screenshot 4. The result is not bad, except for a few artifacts, the most prominent once marked with previews. They are caused by stars that are just on the border of one of the images with slightly different brightness values to the corresponding pixels in the other image. This does not matter a lot for faint structures, but for bright once the differences are huge, especially in a linear image.

Workaround is to stamp out those stars with the CloneStamp tool. Set radius to 14 or so, and strictly speaking you should also set Softness to 0 to get a sharp change to black (I forgot that during my work - it was not harmful). Do GMM again with the modified image. Settings in screenshot 4 (next post).

4. The result in screenshot 5 (next post) no longer has the artifacts - there are some more if you look very carefully, but they can be handled the same way.

Let me know if this procedure solves your problems. It should be the same for the other channels as well. It would be interesting for me to know what you did differently, and why this caused the problems you see.

Georg

[georg.viehoever]

Screenshot 2: Reference mosaic, and Star Alignment settings
Screenshot 3: Alignment results, and GMM 1 settings

[georg.viehoever]

Screenshot 4: GMM1 result with problems marked as previews. CloneStamped image in Preview 1. Clonestamp settings. GMM2 settings (identical to GMM1, except for CloneStamped file).
Screenshot 5: GMM2 result, no longer an artifact in Preview 1

[jeffweiss9]

Georg-
   Thanks tremendously.  That, and your previous response above certainly shows how it should be done to me and others.  In terms of differences, first, I did not check Distortion Correction in either StarAlignment process, nor did I select the 2-D Surface Splines (They weren't selected in Steve Allan's video).  I also, of course, didn't CloneStamp out the bright stars near the border that apparently caused just about all of the problems at least for this Luminosity mosaic since the levels also look pretty good (even though there are more bright stars to clone out). You said above to clone out the bright star in the panel that comes closest to the edge.  Do you mean by that the one on the panel that comes closest to the edge of its own panel ?  Also I wonder why you started with the A5 image in the first StarAlignment step.  Was there a reason you chose that one to start (I did everything in numerical order A1-A5).   A similar question for your GMM process step: Can you explain why you chose the ordering A1-A2-A3-A5-A4.  I suspect you had a reason.  (Again I had just done everything in order).
   I will forge ahead with this new knowledge.  The bigger challenge will be the Blue mosaic.  I suspect I will have to apply DBE to get the levels right, as you did in your previous example, but we'll see.
-Jeff

[georg.viehoever]

Hi Jeff,

just some quick answers before leaving for the daytime job. Maybe they reach you before you go to bed in California...

- the distortion correction was not checked in Steve's video because the feature simply did not exist at this time.
- I mean to clone out the stars on the panel that comes closest to the edge of its own panel.
- I did not chose a specific order when doing the StarAlignment for the reference mosaic. If this turns out to be a problem, I would probably work from inside to outside (to keep distortions low in the center of the image), or would do a hierarchical approach (merge left half, merge right half, merge both halfs) because this reduces the number of necessary StarAlignments., If everything else fails, you could work with a synthetic star field (StarGenerator) or the Plate Solver family of scripts.
- GMM panel order: In Average mode, the order does not matter.

Georg

[papaf]

Georg,
I already saved this topic and your reply in particular in my bookmarks. This is just pure gold. Awesome!
Attached is the result following your tutorial. Mind that this is 1, just one image per channel, 4 panels mosaic.

Thanks again!

[jeffweiss9]

Georg-
  I think it worked great for Luminosity (attached here).  On to R, G, B.
Thanks a million for the great tool and the extras to handle the glitches.
Clear skies,
-Jeff

[georg.viehoever]

Hi Jeff,

great to hear it works for you!

I also had a go at your blue channel. The same basic procedure seems to work there, same kinds of artifacts. Some hints though:

- I had difficulties to get the top left images fit into the reference mosaic. Defining a preview in the mosaic of the previous step helped - this focusses the efforts of StarAlignment on this region.

- Some of your images have artifacts at the border, see top left of screenshot. These did not cause problems for this mosaic, but it did in others: GMM thinks that the transitions to black is part of the image, and you will see the transition in the GMM result as well. My recommendation: remove such borders in the original data (for instance with DynamicCrop). This also avoids potential issues with FrameAdaption.

- To find issues at the transitions between images, the following procedure helps: Ask GMM to generate a mask, use CurvesTransform to make bright parts of the mask less bright, and activate the mask on the merged image. You immediately see where to look for issues, and you can make the mask visible/invisible with PI buttons (see screenshot).

Georg

[jeffweiss9]

Georg-
  The Red and the Green went extremely well also but the Blue still gave me problems -- again in the top left panel. There were a set of star artifacts along the middle of the overlap region that are roughly midway between the edges of A4 and A5 that had to be cloned out.  The results (attached),however, shows the cloned areas with different contrast.  I had set the softness to 0.0 for a hard transition which didn't seem to show this in any of the other filter mosaics, so maybe this is still just a problem with the contrasts not being equalized. 
   I also didn't understand how "Defining a preview in the mosaic of the previous step helped" in focussing StarAlignment on this (upper left ) region.  It certainly is having trouble with that region in my go at it.
   Comments on this appreciated.
-Jeff

[georg.viehoever]

Hi Jeff,

the artifacts you see are surprising: I never saw them before. They usually are only at the border of one of the panels, not in the overlap region. Attached is the result of my GMM run in that region (without any CloneStamping etc) - the artifacts are only where I expected them. My guess would be that StarAlignment for A5 did not work well.

Maybe your A5 suffers from the soft borders issue that I tried to outline in my last post. Cut them away before merging (only strange that this is no issue in my test).

Regarding the Preview in StarAlignment: This process has a setting "Restrict To Previews" in the Star Matching section. It restricts the matching process to the region you marked with a preview (also see the documentation in http://pixinsight.com/doc/tools/StarAlignment/StarAlignment.html#usage_011). After I had created a reference mosaic with A1-A4, I marked the top left part of the mosaic with a Preview, then added A5. Without the Preview, alignment sometimes failed.

Georg

[jeffweiss9]

Georg-
 Your last image (A4-A5) suggests to me that you are creating a separate reference mosaic for each filter (your A5 extends upward further from m32 than mine).   I have been using the same original reference mosaic generated for luminosity for all of the other filters assuming these need to overlay precisely on the luminosity.  Is that my mistake?
-Jeff

[georg.viehoever]

Yes,maybe thats the issue. I did make a second reference mosaic for the blue channel.