Back to forums

Mosaic Question

Processing Mosaic Deep Sky
23 replies•387 views
Mark McComiskey avatar
Mark McComiskey avatar
A pick your brain question. 

I have not done a mosaic before, but am starting to think about doing so.

Boiled down, as I understand it, you end up stitching together two masters (lets assume a two-panel mosaic) . 

When taking the subs for each master, you specify an area of overlap.  So in theory, assuming equal integration time for each panel of the mosaic, the raw data across all subs would have twice the signal in the area of overlap than in the areas without overlap. 


When the masters get fused together, does the area of overlap benefit from the cumulative integration time accumulated in that area across all subs?  Or does the fact that they are being added as masters mean that one or the other is being picked and there is no cumulative overlap benefit?  I suspect the latter from what I have read, but that seems such a waste.  If that were the case, I would be inclined to integrate the area of overlap separately from the non overlapping areas and then attempt to stitch them together.

I am sure the answer is obvious to people who have been at this longer than I have, but would appreciate any insight.
Well Written Respectful Engaging
danieldh206 avatar
danieldh206 avatar
The overlap will get blended. Astro Pixel Processor is the easiest way to stitch the masters together. You will stitch together the linear images.
Oscar avatar
Oscar avatar
this is the way I've done it and it's basically just registration with some extra steps (no noticeable overlap, or seam, if you do it right, and get an equal amount of integration for all panels) https://www.galactic-hunter.com/post/pixinsight-mosaic
andrea tasselli avatar
andrea tasselli avatar
Mark McComiskey:
A pick your brain question. 

I have not done a mosaic before, but am starting to think about doing so.

Boiled down, as I understand it, you end up stitching together two masters (lets assume a two-panel mosaic) . 

When taking the subs for each master, you specify an area of overlap.  So in theory, assuming equal integration time for each panel of the mosaic, the raw data across all subs would have twice the signal in the area of overlap than in the areas without overlap. 


When the masters get fused together, does the area of overlap benefit from the cumulative integration time accumulated in that area across all subs?  Or does the fact that they are being added as masters mean that one or the other is being picked and there is no cumulative overlap benefit?  I suspect the latter from what I have read, but that seems such a waste.  If that were the case, I would be inclined to integrate the area of overlap separately from the non overlapping areas and then attempt to stitch them together.

I am sure the answer is obvious to people who have been at this longer than I have, but would appreciate any insight.

*In general it doesn't work that way as the overlapping regions  wouldn't have the same gradient and signal distribution so that to create a seamless mosaic you would need to feather on side so to match the other side and thus using more overlap than average. In few instances you can plainly average the overlapping sections and still end up with a seamless mosaic too but I can't remember many instances of this happening, especially with larger mosaics. Besides, economhy would dictated the overlapping sections been kept to a minimum so no real advantage can be had by just averaging the matching regions (without even considering the obvious issue of what happens outside those regions).
Mark McComiskey avatar
Mark McComiskey avatar
Okay.  I am gathering from this that despite the fact that twice as much data is collected for the overlap region as for the non-overlap regions, that extra signal is essential lost in the mosaiac process.  Pity.
andrea tasselli avatar
andrea tasselli avatar
Signal to noise ratio adds in quadrature and for deep enough images (the ones you really want) the effect is marginal.
Mark McComiskey avatar
Mark McComiskey avatar
Twice the signal is a 1.4x improvement in SNR. If I could get a 40% improvement in my SNR, in particular when it would essentially be “free”, I would happily take it.
Well Written
Dean Ostergaard avatar
Dean Ostergaard avatar
Mark McComiskey:
Okay.  I am gathering from this that despite the fact that twice as much data is collected for the overlap region as for the non-overlap regions, that extra signal is essential lost in the mosaiac process.  Pity.

Why is that a "pity"? What in the world would you do with an image that had a 10% wide strip of image that is obviously higher quality than the rest of the image?

Unless you only plan to keep that 10%? That makes no sense.

If anything you want the overlap in a mosaic to be indistinguishable from the rest of the image. IOW, it shouldn't look like a mosaic.
Helpful Insightful
Mark McComiskey avatar
Mark McComiskey avatar
“That makes no sense” is somewhat harsh. 

What would I do with it?  Given the framing I am looking at for the object in question, I would hope that it would allow me to bring out more of the faint structure that sits largely in the area of overlap than in the non overlapping regions. 

I suppose it may well be the case that the higher SNR area would look jarring vs the lower SNR areas, but perhaps not. I suspect the degree of distinction between the two areas would be somewhat dependent on what was in the frame. 

As it is, given how the mosaic process works, I think I will end up integrating the area of overlap separately and experiment with trying to blend in any areas of interest from the higher SNR section.
andrea tasselli avatar
andrea tasselli avatar
Mark McComiskey:
Twice the signal is a 1.4x improvement in SNR. If I could get a 40% improvement in my SNR, in particular when it would essentially be “free”, I would happily take it.


*In the highly hypothetical case of having equal SNR. In the most likely case of uneven quantities that is less than the square root of 2. In my extensive iterations with various methods of mosaicking the areas of overlap might, on average, have a slightly better SNR than the panes adjoining them whichever method you use to blend them together but in the economy of a mosaic, if this is what you are trying to do and not quite sure you do, this hardly ever matters.
Alan Brunelle avatar
Alan Brunelle avatar
Mark McComiskey:
Twice the signal is a 1.4x improvement in SNR. If I could get a 40% improvement in my SNR, in particular when it would essentially be “free”, I would happily take it.

*If you want to learn the how and why's of mosaicing, just do it by your simplistic method.  And then use traditional processes. It's not like you are going to destroy the masters. 

If you have collected just two panels with, for example, 15% overlap, do a registration for mosaic.  Then do a simple addition of the two frames.  You'll end up with some misalignment issues, but ignore that for the exercise.  You will find that the overlap will have the slightly improved S/N.  Now, have fun trying to process that. Not so fun.  Then choose another method of mosaicing and learn first hand why it is done the way it is first hand. 

Now try collecting a third overlapping frame and repeating. You will then learn that you now have something like 4 different regions of S/N to deal with and if you crop and discard the two outer, lower S/N non-overlapoed sections, you will end up with a mosaic that is no larger than a single frame!  Which you would have achieved if you would have centered your field properly to begin with!

Bottom line, yes mosaicing, by definition, requires that there is a small amount of sacrifice of information to achieve a realistic wider field, at best.  The solution is that you collect enough minimal data for each panel to achieve your required result in the first place and then not worry about the improved S/N in the overlap.  If you were strict about holding onto that info in the overlap, you would have an image with a stripe down the middle.  If that is your desire, then go for it.

But as I suggested, challenging established methods is a great way to learn how and why something is done the way it is. Occasionally, rarely, doing so will result in a new method.  Even more rare is that new method is better. But if you come up with a new way, please let us know!
Mark McComiskey avatar
Mark McComiskey avatar
Will do.
Alan Brunelle avatar
Alan Brunelle avatar
Mark McComiskey:
“That makes no sense” is somewhat harsh. 

What would I do with it?  Given the framing I am looking at for the object in question, I would hope that it would allow me to bring out more of the faint structure that sits largely in the area of overlap than in the non overlapping regions. 

I suppose it may well be the case that the higher SNR area would look jarring vs the lower SNR areas, but perhaps not. I suspect the degree of distinction between the two areas would be somewhat dependent on what was in the frame. 

As it is, given how the mosaic process works, I think I will end up integrating the area of overlap separately and experiment with trying to blend in any areas of interest from the higher SNR section.

*I think the issue here is your first question left out information about what you are trying to achieve.  In a sense, you are wanting to place the image of a rather smaller object into a wide field of less importance.  Fact is there really is no reason why you cannot just plop an object into a field (and here I mean accurately aligned) from even two different data sets. Even from two different optics and cameras. See my

Intergalactic Travelers' View of Abell 347, From NGC891 to NGC1003
Where I upscaled a lower resolution wide field image of a field and then added a higher resolution image of a very limited field including a galaxy.  Maybe a pointless exercise, but I learned something!  Read my description and look at the revision.
Alan Brunelle avatar
Alan Brunelle avatar
Alan Brunelle:
Mark McComiskey:
“That makes no sense” is somewhat harsh. 

What would I do with it?  Given the framing I am looking at for the object in question, I would hope that it would allow me to bring out more of the faint structure that sits largely in the area of overlap than in the non overlapping regions. 

I suppose it may well be the case that the higher SNR area would look jarring vs the lower SNR areas, but perhaps not. I suspect the degree of distinction between the two areas would be somewhat dependent on what was in the frame. 

As it is, given how the mosaic process works, I think I will end up integrating the area of overlap separately and experiment with trying to blend in any areas of interest from the higher SNR section.

*I think the issue here is your first question left out information about what you are trying to achieve.  In a sense, you are wanting to place the image of a rather smaller object into a wide field of less importance.  Fact is there really is no reason why you cannot just plop an object into a field (and here I mean accurately aligned) from even two different data sets. Even from two different optics and cameras. See my

Intergalactic Travelers' View of Abell 347, From NGC891 to NGC1003
Where I upscaled a lower resolution wide field image of a field and then added a higher resolution image of a very limited field including a galaxy.  Maybe a pointless exercise, but I learned something!  Read my description and look at the revision.

*Mark,

Sorry, but the link in that I gave you above no longer has the raw stitched field that I thought I had put in the revisions of the field.  I don't know why I removed it from that post because it shows a bit more of what I was talking about in my description.  Below is a heavily size reduced image to convey what I was talking about.



Hope that helps!

Alan
Mark McComiskey avatar
Mark McComiskey avatar
Fair. I should have outlined why I was asking this question. Extremely faint gas clouds between two bright objects. The bright objects would be outside the overlap area, the faint gas clouds within.
andrea tasselli avatar
andrea tasselli avatar
Contain the feathering areas to few pixels either way and averaging in the overlapping areas would get you there, assuming PI, well flattned masters and PhotometricMosaic usage.
Tony Gondola avatar
Tony Gondola avatar
Mark McComiskey:
Okay.  I am gathering from this that despite the fact that twice as much data is collected for the overlap region as for the non-overlap regions, that extra signal is essential lost in the mosaiac process.  Pity.

If you want more signal increase integration time. Mosaicking is not the way to do it. If you think about it, doing it that way ruins the mosaic.
Dean Ostergaard avatar
Dean Ostergaard avatar
Mark McComiskey:
“That makes no sense” is somewhat harsh.


Sorry. Others have since said more eloquently what I was pointing out that enhancing the seams of a mosaic defeats the usual goal of creating a seamless image from many smaller ones.

@Alan Brunelle has given you the best answer to what you are trying to achieve. Unfortunately, there are no shortcuts to higher quality images. Well, except maybe more money. 
Tony Gondola avatar
Tony Gondola avatar
Dean Ostergaard:
Mark McComiskey:
“That makes no sense” is somewhat harsh.


Sorry. Others have since said more eloquently what I was pointing out that enhancing the seams of a mosaic defeats the usual goal of creating a seamless image from many smaller ones.

@Alan Brunelle has given you the best answer to what you are trying to achieve. Unfortunately, there are no shortcuts to higher quality images. Well, except maybe more money. 

And more time....
GiffS avatar
GiffS avatar
Having just finished a 9-Panel Mosaic of The Markarian's Chain region I can safely tell you that any hope of getting "extra integration" time in the overlap is  the least of your worries.  When you start looking at bigger mosaics you quickly learn that getting anything close to normal integration time on each panel is a challenge. To get 16 subs of 300 seconds each took two full nights of imaging time. At this time of year where I am that's somewhere around 7 possible hours a night. So basically two nights of telescope time resulted in an image with just 1.3 hours of integration time. Larger mosaics are a serious commitment of time and resources and that's with a OSC camera. I couldn't imagine doing it in mono.  

What I learned is that the juice had better be worth the squeeze. In my case I don't think it was. Next time I will plan for a large nebulous region in the winter when I have lot's of dark time per night and a more detailed target. Here's a link to the mosaic:

https://app.astrobin.com/i/m9hbu2
Well Written Helpful Insightful Engaging
Tony Gondola avatar
Tony Gondola avatar
I think another aspect of mosaics that's often overlooked is even if you do it, how are you going to present it? Depending on the size of the image, even an 8K screen won't show the image at full resolution. a four panel mosaic with an APS-C sensor will greatly exceed that. I even see that now with single images from large sensors displayed on common screens. Much resolution is lost on a regular screen unless you zoom in and pixel peep. If you are going to do it, making a large digital print at full resolution is one of the few end uses that makes any sense to me.
Helpful Insightful Respectful Concise
Mark McComiskey avatar
Mark McComiskey avatar
I agree with all those downsides of mosaics. 

I am really just contemplating a 2 panel mosaic for a specific target on which, to get the framing I would like, I need perhaps 25% more dimension that I have natively. 

I’ve seen some of the 8-10 panel mosaics out there, and my hat goes off to them, but I have no inclination to try that. I would likely just use a shorter focal length for something like that.
Well Written
Tony Gondola avatar
Tony Gondola avatar
Mark McComiskey:
I agree with all those downsides of mosaics. 

I am really just contemplating a 2 panel mosaic for a specific target on which, to get the framing I would like, I need perhaps 25% more dimension that I have natively. 

I’ve seen some of the 8-10 panel mosaics out there, and my hat goes off to them, but I have no inclination to try that. I would likely just use a shorter focal length for something like that.

I think that might be one of the better end uses for the technique.
Brian Boyle avatar
Brian Boyle avatar
Mark McComiskey:
I agree with all those downsides of mosaics. 

I am really just contemplating a 2 panel mosaic for a specific target on which, to get the framing I would like, I need perhaps 25% more dimension that I have natively. 

I’ve seen some of the 8-10 panel mosaics out there, and my hat goes off to them, but I have no inclination to try that. I would likely just use a shorter focal length for something like that.



Yes.  I think this is the best way to think of mosaics.  I might even extend that to a 3 x 1 mosaic.  And you can get really good results very easily this way.  But as soon as you have multiple row/columns the task gets really tricky.  Abandon hope all who enter here.

For small linear mosaics I have found that the PI AlignByCoordinates/TrimMosaic/PhotometricMosaic path to work pretty well, if a little tedious. GradientMergeMosaic can work even better, but in very rare circumstances.  In general APP does much better that PI for complex mosaics, but I don;'t see a meed to use that for the simplest mosaics.

Both APP and PI struggle with astrometric solutions for mosaics larger than 50degrees in extent e.g. stitching together short focal-length images.  For that I would suggest PTGui [better than PS for mosaicing].  


CS Brian
Helpful Insightful