Help Requested w/Gradient Merge Mosaic Problems

[kerrywaz1]

I have a 9-frame image I have been trying to process using Georg's Gradient Merge Mosaic module. I followed Catanonia's video to align the frames etc. However, when I combine the frames using the Gradient Merge Mosaic module as depicted in the video, I get these strange "burned out" spots around some of the bright stars.

I have tried different combinations of the parameters, and the attached image has as good as I could get. For this particular try, I set "Type of Combination"="Average", "Shrink Radius"=1 and "Feather Radius" = 3.

All of my frames are linear with only dynamic cropping and DBE applied. As can be seen, the gradient merge is not working all that well either. This is probably due to some of the frames having high clouds and/or light pollution issues.

Can someone please advise on what I might try to clean this mosaic up?

Thanks!

[georg.viehoever]

Kerry,

those star artefact usually happen when bright stars cross the borders between tiles. The work around is to drastically increase the feather radius (>20). There are other workarounds, but you first should try this.

To remove the gradients between tiles, increasing the shrink radius, maybe to 5. This should remove gradients at the border of your tiles that GMM cannot remove if they are there.

Georg

[kerrywaz1]

Hi Georg,

Thanks for the reply.

I played with the settings again as you suggested and the settings of shrink radius = 8 and feather radius = 21 gave the attached result. While the dark star artifacts are somewhat reduced, I could not get them to disappear completely. All the other setting combinations I tried made were not as good.

What else can I try?

Thanks

[georg.viehoever]

Except for the remaining star artefacts the seams are now gone, so GMM is now doing  a pretty good job. What it apparently also cannot compensate are the large scale brightness differences your panels have. You need to work on these.

I would like to experiment with your data. Could you provide FITS files of your panels for download, for example the top right and center right one?

Georg

[kerrywaz1]

Hi Georg,

Any thoughts on how I might get those brighter tiles to match up?

I am sending the FITS files to you in a PM.

Thanks for looking at this,
--- Kerry

[georg.viehoever]

Kerry,

I did not have much of a problem with your data. I think the key issue here is that your panels have fairly different statistics/basic brightness. To merge them successfully, you need to adapt the brightness to each other, and fortunately it is very simple with PI. Basically, I have been following the procedure outlines in Steve's video on http://www.youtube.com/watch?v=H2r3ZYqCaP8&gl=US&lr=1:

• Create Master Mosaic (see first screenshot). Note that FrameAdaption is checked here. The resulting mosaic is already almost seamless
• Align the individual frames (screenshot 2). Note that FrameAdaption is checked here.
• Do GMM with the aligned frames (screenshot 3) with default settings. Perfect result.

Hope that helps.
Georg

[kerrywaz1]

Hi Georg,

I followed Steve's video as well and it appears there is an order to building the reference mosaic and how well the panels integrate. Regardless, I am still having trouble with the star artifacts in GMM (see attached).

I uploaded all of the rest if the 9 frames to DropBox if you care to take a look and see if you can get it to work better. I will send you the links in a PM.

Thanks,
Kerry

[georg.viehoever]

Ok, will have a look later this week.
Georg

[kerrywaz1]

Ok, thanks.

BTW, is there a way in PI outside of the Star Alignment tool to get those panels to match better in brightness?

--- Kerry

[georg.viehoever]

...BTW, is there a way in PI outside of the Star Alignment tool to get those panels to match better in brightness?

LinearFit process

[kerrywaz1]

Georg,

I haven't used Linear Fit before and would like to see how it works with this dataset.  Also, I thought Linear Fit only worked with stretched non-linear images? 

--- Kerry

[georg.viehoever]

I haven't used Linear Fit before and would like to see how it works with this dataset.  Also, I thought Linear Fit only worked with stretched non-linear images? 

Kerry,

Linear fit essentially does the same a "frame adaption" in the StarAlignment tool: Linearly adjust one image such that identical locations of two images have minimal differences. It certainly is best applied to linear images. See http://pixinsight.com/forum/index.php?topic=2302.msg15204#msg15204.

Georg

[georg.viehoever]

...I uploaded all of the rest if the 9 frames to DropBox if you care to take a look and see if you can get it to work better. I will send you the links in a PM.
...

Hi Kerry,

I am having a look at the images. It will take some time until I have final results, because the resulting 9695x6845 pixels are hitting the limits of what I can do on my 4GB laptop. Let me start by introducing the baseline result:

• the data is for a 9 panel mosaic, with the data shown in screenshot 1. The subimages are shown in the approximate locations of the final mosaic. Special is only the RightCntr frame: It needs a 180 degree rotation to fit to the central panel.
• The data can be combined into a master mosaic that is later used for aligning the subpanels as shown in Steve's video. Frame adaption has been activated. The order of combining the panels has been
  527
  418
  639
  I had to define a preview in step 4 for the LeftCntr panel (as shown in screenshot 1) for StarAlignment to work because of the small overlap to MiddleCntr. Plus RightCntr (step 8 ) took a long time to succeed (in try 21 of StarAlignment). Otherwise no problems. The resulting mosaic is shown in screenshot 2. We clearly see that frame adaption has not been able to adjust for the differences between the panels: usually just one of the neighbors matches well, the others show clear seams.
• Running GMM with defaults produces screenshot 3. We see that it smoothed out the seams, but we see star artefacts, and the large area differences also did not entirely disappear.

We will discuss the analysis and remedies for those issues in later posts.

Georg

[georg.viehoever]

Screenshots 2+3

[georg.viehoever]

Now lets attack the first problem, the star artefacts. They are generated if the brightness information (or more exactly: the gradients of the brightness) differ wildly between images in the regions close to the border of one image. In trying to find a compromise between the conflicting information, GMM generates those artefacts. Prime candidates are bright stars that have different brightness values (for whatever reason) in the different panels. Usually, those differences can be smoothed out by increasing the Feather radius parameter of GMM. But if the differences are too large, this does not help.

The most prominent of these artefacts can be seen at a bright star that is share between LeftCntr and LeftBottom. Screenshot 1 nicely shows its location relative to the different panels. For visualization, I added the mask generated by GMM to those images, so the red overlay shows the limits of the panels. In panel LeftCntr this star has a brightness of 0.86, in LeftBottom 0.75. This is simply too much for GMM.

How to fix this? Remove the conflicting information ! Use clone stamp to remove the critical star in the panel where it is closest to the border (screenshot 2). Note the use of Radius=1 and Softness=0 to avoid aliasing effects. I did this for about 10 locations within the mosaic, ran GMM with default parameters, and got the result in screenshot 3.

The resulting mosaic has perfect seams, which is just what GMM should do.  But the different panels can still be due to large scale brightness differences between them. I plan to discuss reasons and remedies for the issues in future posts.

Georg