GradientsMergeMosaic sufficient overlap?

[pfile]

how much overlap is needed between mosaic panes for GMM to do it's magic?

i currently have 675 pixels of overlap (which amounts to roughly 20% of the width of my sensor). when planning the mosaic i didn't really think about the field curvature of my RC; despite using a flattener the stars are distorted near the edges, so i'd like to crop off as much as i can get away with.

the mosaic is looking pretty good, but i'd just like to optimize it a little bit. there's one pane where the stars in the vicinity of the seam did not line up properly, and i'm hoping that cropping it a little more is at least part of the solution.

thanks

rob

[georg.viehoever]

The required overlap is ShrinkRadius+FeatherRadius+ (1 or 2) pixels. More overlap is neither necessary nor does it change the quality.

[Zocky]

The required overlap is ShrinkRadius+FeatherRadius. More overlap is neither necessary nor does it change the quality.

Can you please put that in percentage? Or give an example how to calculate for sensor with 3352x2532 px?
Thnx.

[pfile]

thanks - so it's just a handful of pixels then? cool.

that's helpful as i should be able to crop a whole lot off of each image.

[georg.viehoever]

The required overlap is ShrinkRadius+FeatherRadius. More overlap is neither necessary nor does it change the quality.

Can you please put that in percentage? Or give an example how to calculate for sensor with 3352x2532 px?
Thnx.

if ShrinkRadius is 1, FeatherRadius is 10, your images should at least overlap 11 pixels (maybe one or two more, but not much more). So for your 3352x2532, the overlap should be 11 pixels, or 11/3352*100%=0.03% in x. I'll leave the  y-direction as an exercise for you  .
Georg

[Zocky]

Thnx Georg

[pfile]

so i guess a related question is when StarAlignment is doing frame adaptation, does it need a larger overlap?

or does SA need a large overlap simply to compute the proper alignment in the face of distortions?

in other words we still probably have to stick to 15-20% overlap during panel aquisition...?

rob

[georg.viehoever]

I am not sure about frame adaption. But since this is a 1 dimensional linear regression between corresponding non-zero pixels, you need a sufficient set of pixels to do this with sufficient accuracy. I would think that 10% of all pixels is more than sufficient.
Georg

[jerryyyyy]

These weremy questions on star alignment... which is the first step, right?  I was hoping to make a mosaic of M95 and M96, which are not too far aparert, but I cannot get the star alignment to work.  I wonder if there are optimal settings to maximize what overlap you have... humm I guess you could use the dynamic version and pick out the stars. 

[georg.viehoever]

These weremy questions on star alignment... which is the first step, right?  I was hoping to make a mosaic of M95 and M96, which are not too far aparert, but I cannot get the star alignment to work.  I wonder if there are optimal settings to maximize what overlap you have... humm I guess you could use the dynamic version and pick out the stars.

StarAlignment itself only needs very small regions of overlap. Please read its documentation for some of the tricks. If these fail, you can always do DynamicAlignment.
Georg

[Andres.Pozo]

When there is little overlap between the tiles you can solve it by different means:

• Use a synthetic image that covers the full mosaic. You can generate it with the script CatalogStarGenerator. I would recommend doing this even if there would be enough overlapping since this way the aligning errors don't accumulate after applying many times StarAlignment.
• You can use the new script MosaicByCoordinates that doesn't require any overlap for aligning the tiles and also has other capabilities. You can find it in this thread: http://pixinsight.com/forum/index.php?topic=6535.0.
  This script is still a beta although it should work OK.
  

[pfile]

your script works great, i had tried it on a 1x2 mosaic that i shot last night but i will go back and try it with my 2x3 that i have been working on (that was the subject of this thread)

rob

[jerryyyyy]

These weremy questions on star alignment... which is the first step, right?  I was hoping to make a mosaic of M95 and M96, which are not too far aparert, but I cannot get the star alignment to work.  I wonder if there are optimal settings to maximize what overlap you have... humm I guess you could use the dynamic version and pick out the stars.

StarAlignment itself only needs very small regions of overlap. Please read its documentation for some of the tricks. If these fail, you can always do DynamicAlignment.
Georg

Yes, my problem was I cropped too much.  When I used the uncropped images they aligned.  Still need more subs to make the mosaic work....

[jerryyyyy]

I am still haveing problems merging.  After I get my images of M95 and M96 merged I get an Alignment image but when I try to run the next step on the two original images I get an error on only the M96 image but not the M95. 

This is the error... seems like there are plenty of stars for the match. 

ImageIdentifier: Processing view: M96_LRGB_mosaic
id = Align
0.0100 s

StarAlignment: Processing view: M96_LRGB
Align:
Structure map: done
Detecting stars: done
449 stars found.
M96_LRGB:
Structure map: done
Detecting stars: done
321 stars found.
* Distortion correction: Iteration 1 of 20
Matching stars: done
123 putative star pair matches.
Performing RANSAC: done
* Previous attempt failed - this is try #2
useScaleDifferences=true
* Distortion correction: Iteration 1 of 20
Matching stars: done
101 putative star pair matches.
Performing RANSAC: done
* Previous attempt failed - this is try #3

goes on and on until it terminates. 

I wonder if I just need better images. 

[pfile]

did you try making previews over where the images should overlap?

also you can turn on "detected stars" mode to see just what is being detected. could be hot pixels in which case you'll never get a result. also if your star shape are soft you may need to increase the star profile parameter in StarAlignment

rob