Distorted Milky Way Mosaic

I've given this more thought and I think I can guess what's happening.  Behind the scenes, the individual panes of the mosaic are being mapped onto a sphere - this is entirely correct.  Then the sphere is projected onto a flat plane to produce the output image.  For the equirectangular projection it can be imagined as a projection of a map of the earth where longitude becomes the horizontal direction and latitude becomes the vertical direction.  My bottom pane is the reference image and so it sits on the "equator" and then I naturally see increasing horizontal stretch for the next panes because of the projection.

The kind of projection I really need for a Milky Way panorama is a Cylindrical Projection. 

Maybe I can achieve the effect I want by rotating each image through 90 degrees before opening them in APP so all the panes sit on the Equator?

Mark

Unfortunately, rotating each image through 90 degrees before opening them in APP didn't work either.  Here is the Equirectangular result:

Oddly, it resembles a Rectilinear projection but you can see from the auto-generated filename that I did definitely choose Equirectangular.

Mark

Hi Mark -

I had tried a similar project (very wide field mosaic of about 180 degrees) around last May.   I had similar issues when trying to create the mosaic in APP.  As you noted, I was able to accomplish it successfully using a cylindrical projection (using Photoshop's automated photomerge).  I had posted this information back then, and I believe that Mabula had added to his development list to include cylindrical projection in APP.  So, to my knowledge, it's a feature that will be there at some later date, but is not available at this time.

Scott -  http://www.astronomersdoitinthedark.com/

Problem Solved!!

I went through all the tool tips that might be relevant and under registration model on the "4) REGISTER" tab I found the following hint: "The calibrated projective model is required if your FOV is greater than 120 degrees, especially in mosaic mode".

So I switched to "calibrated projective" and rotated my exposures so the result would be a horizontal (not vertical) ribbon and here's the result without any other processing:

If you try to build a vertical Milky Way it still distorts:

I'm so pleased because this data is 2 years old and the thought of trying to align the pieces has filled me with dread.  I can now go ahead and properly stack the pieces knowing that APP can align it all with incredible precision!

Mark

Posted by: Mark Shelley

I have done a very quick 3 pane mosaic of the Milky Way using an old Nikon 35mm lens on a Sony A7S camera.  The 3 frames register precisely but the final integrated image has crazy amounts of distortion.  It hasn't come out as I expected, so I'm wondering if I've done something wrong?  I deliberately chose the bottom frame as the reference frame to emphasize the issue 😉

Here's the result of the EquirectangularProjection:

 

And here's the result of the RectilinearProjection:

Mark

Hi Mark,

Yes, this is all about how the data is projected.

First let me clarify the following: the equirectangular projection is only possible if you registered with the calibrated projective model. I already read that you found this as well in the registration tooltips 😉

The reason is that to be able to project the data differently, the software needs to know parameters like

• the focal length at which the exposures were shot,
• and the principal point, that's the location on the sensor where the optical axis hits it.

These are found using the calibrated projective mode

In the rectilinear projection, the distortion that we see is perspective distortion, the larger the field of view becomes, the more the data is stretched. (This projection keeps straight lines straight.) From that image I do also see that the registration is fine, no strange distortions appear.

In the equirectangular projection, the data is indeed mapped on a sphere and than projected on a plane with longitude and latitude coordinates. Currently the implementation is such that you have 360 degree horizontally and 180 degrees vertically. That is the reason for the equirectangular projection to look a bit strange, especially since the middle of the reference frame has the coordinates of [0,0], or is treated as the equator. So if you choose the middle frame as you reference, the projection will look better.

Mabula

Posted by: Mark Shelley

I've given this more thought and I think I can guess what's happening.  Behind the scenes, the individual panes of the mosaic are being mapped onto a sphere - this is entirely correct.  Then the sphere is projected onto a flat plane to produce the output image.  For the equirectangular projection it can be imagined as a projection of a map of the earth where longitude becomes the horizontal direction and latitude becomes the vertical direction.  My bottom pane is the reference image and so it sits on the "equator" and then I naturally see increasing horizontal stretch for the next panes because of the projection.

The kind of projection I really need for a Milky Way panorama is a Cylindrical Projection. 

Maybe I can achieve the effect I want by rotating each image through 90 degrees before opening them in APP so all the panes sit on the Equator?

Mark

Excellent, that indeed is what is happening 😉 , a vertical cylindrical projection would suite the data most, I agree.

Posted by: Mark Shelley

Unfortunately, rotating each image through 90 degrees before opening them in APP didn't work either.  Here is the Equirectangular result:

Oddly, it resembles a Rectilinear projection but you can see from the auto-generated filename that I did definitely choose Equirectangular.

Mark

 

I think it does show the rectilinear projection, maybe there's a bug in the integration file naming here. Need to check that.. 😉 the equireactngular projection is only possible with the calibrated projective registration mode and so the GUI should only then show it as an option ideally...

Mabula

Posted by: Scott Rosen

Hi Mark -

I had tried a similar project (very wide field mosaic of about 180 degrees) around last May.   I had similar issues when trying to create the mosaic in APP.  As you noted, I was able to accomplish it successfully using a cylindrical projection (using Photoshop's automated photomerge).  I had posted this information back then, and I believe that Mabula had added to his development list to include cylindrical projection in APP.  So, to my knowledge, it's a feature that will be there at some later date, but is not available at this time.

Scott -  http://www.astronomersdoitinthedark.com/

Hi Scott & Mark,

Yes, it's on the priorities list ! Other projections will come 😉

First however, the mosaic algorithm for the calibrated projective model needs to get a big upgrade! It isn't using yet the far more advanced mathematical formulation that is now used in the uncalibrated projective mode for mosaicing with same camera and optics disabled.

All these registration settings have different algorithms behind them for the calculation of a mosaic. So after having upgraded all mosaic algorithms, the other projections will come..

Mabula

Posted by: Mark Shelley

Problem Solved!!

I went through all the tool tips that might be relevant and under registration model on the "4) REGISTER" tab I found the following hint: "The calibrated projective model is required if your FOV is greater than 120 degrees, especially in mosaic mode".

So I switched to "calibrated projective" and rotated my exposures so the result would be a horizontal (not vertical) ribbon and here's the result without any other processing:

 

If you try to build a vertical Milky Way it still distorts:

 

I'm so pleased because this data is 2 years old and the thought of trying to align the pieces has filled me with dread.  I can now go ahead and properly stack the pieces knowing that APP can align it all with incredible precision!

Mark

Excellent, you are quite the problem solver Mark 😉 !

Did you get good registration RMS for the whole mosaic?

Be aware that the calibrated projective mosaic mode is up for a big upgrade:

First however, the mosaic algorithm for the calibrated projective model needs to get a big upgrade! It isn't using yet the far more advanced mathematical formulation that is now used in the uncalibrated projective mode for mosaicing with same camera and optics disabled.

I know that if you start to create a mosaic with more than 10 panels, with the current mosaic algorithm in the calibrated projective mode, you will start to see some errors. That will be fixed in the upgraded version though 😉

Mabula

Did you get good registration RMS for the whole mosaic?

 

Approximately 0.22 pixels RMS error across all the overlaps.  That's astonishingly good!

Mark

😉 excellent