The Astro Pixel Processor Group. The group is open for everybody, downloads are available: Download APP.
Large Scale Mosaics
March 29, 2017 at 20:12 #13479
Here is an initial 16 pane mosaic. I used files created in Pixinsight. This time I used ones that I had performed the DBE on, so the gradients were removed, but kept the files in Linear mode.
The joining looks really good, except for two areas where the stars are streaked. 2 images attached to test.
I m going to try a 6×6 full frame stitch later.
March 29, 2017 at 20:13 #13480March 29, 2017 at 20:14 #13482March 29, 2017 at 20:18 #13487
- 1 person likes this.
Hi Tom, you can use the big blue photo button for uploading images (attachment (no Photos)) won’t preview images.
March 29, 2017 at 20:35 #13496
- 1 person likes this.
Okay, you are starting off with big mosaics already 😉
First thing we need to know…
Do you remember how you registered the individual panels in Pixinsight?
If you used any form of thin plate splines, chances are, PI has done more damage to your data than good. So we need to know this before anything else.
Explanation being: the thin plate splines are a terrible solution to correct optical distortion, they don’t correct the reference and the other frames are distorted just to fit to that uncorrected reference frame. PI calls it distortion correction, when it actually isn’t…
Real distortion correction corrects all your data (including the reference) to the rectilinear projection it is supposed to have 😉
So have you used thin plates splines in PI? If you did, it’s possible better of we take another approach.March 29, 2017 at 20:41 #13498
I have scanned the 16 panel, but I am having a hard time to find the 2 error zone you mention.
If you tried this in Registar, would you have to register panel by panel to get the 16 panels stitched together?
Also good to know, what kind of optics were used and what focal length was used?
Is there data from different camera’s and OTA’s ?March 29, 2017 at 23:54 #13502
Here is the 25 pane version, used with the DBE files. Very good blending result for brightness, gradients and detail.
Two areas of overlapping show star distortion. Whether these can be joined without issues by other settings I m noy sure yet. More testing to do.
This took 3 hrs on the laptop to put together.
March 30, 2017 at 00:20 #13509
- 4 people like this.
Wow! You’re putting APP to the test 😉
If the original Panels are affected by thin plate spline transformations, it would be hard to fix, since then the distortions aren’t scientifically exact anymore. The will deviate unlogically from the rectilinear projection due to the thin plates.
But, there are several options that you can try to get a better results
at 3) Analyse stars,
- set minimum star size to 5-6 and or increase clip star profile to 0.3-0.5, this will help with star lokation precision
- increase the amount of stars limit from 2500 to 5000
- lower the detect above noise kappa value from 8 to 4 (below 3 is unadvisable)
You can enable filter star profile, this could also help in increasing the star lokation precision which could help in getting a better result
I notice some frames have very little overlap right?
When frames have very little overlap, it can help to use triangles instead of quadrilaterals in
pattern recognition descriptor shape
That should help in getting those frames that have very little overlap to have better registration.
Let me know if you can improve the result 😉
MabulaMarch 30, 2017 at 11:39 #13517
No option to make new lights and calibration frames using only APP?March 30, 2017 at 20:40 #13545
Yes I like to get straight in there and test the big mosaics ?
I don’t quite understand the Thin Plate Spline bit. That sounds like a method to calculate the registration for the panels.
So let me mention where the frames came from
1) FITs were calibrated in AstroArt 5 with Darks and Flats applied here to stack the master 4 hr files.
2) Once I had the master frames for each panel, I opened them in PI, Cropped any border from the stack, then ran the DBE process to remove any gradients and vignetting. (I have all 25 frames without the DBE which I can mosaic in APP to test v the DBE version)
3) So PI has not registered these panels to each other. They have only had the DBE gradient removal performed on them.
There is approx. 15-20% overlap on the frames. Or at least I thought there was. The gap on the bottom row surprised me, the rotation effects around Polaris is evident as it seems to compress the top row frames. I will re-run the 25 with your suggestions and see how I get on.
March 30, 2017 at 20:48 #13546
- 1 person likes this.
My method for my two Mega Mosaics, Orion and the Galactic Centre were to make a base layer in PI with their Star Alignment, process. The GMM process which can make a mosaic from the Star Alignment base layer did not work for me. It left halos on stars, dark holes, distortions etc.
So I let the SA process register and build the base layer. Then I used the base layer from this in Registar. I would let Registar align a non linear panel to the overall base layer. Then I would build the mosaic in Photoshop, and blend them in.
This is data from 3 Takahashi FSQs (0.53m), 2 Atiks and 1 SXV ccd camera. So almost all the same set up. Pixel size is different in the SXV camera.
Very time consuming, but it allowed me to erase some overlapping regions which had bad registered stars. I do like having the full 25 pane layer in Linear Mode. I could re-blend in some panels to patch the distorted stars if necessary. However first a re run with your suggested settings.
March 31, 2017 at 00:41 #13562
- 2 people like this.
Hi Tom, let me know if you get more improvement with my suggestions 😉
The mosaic algortihms could have some more fine tuning definitely, so i’ll be happy to test using your data to see if we can improve the results 😉
The number of parameters APP needs to calculate for such a 25 panel are quite high, about 25 * 12 = 300
So 300 parameters need to be solved for several 1000s of stars.
But I know I can complicate the models further and possibly improve more.
About the thin plates: I was under the impression that your mosaic panels are previous integrations from PI. Aren’t they? If not, forget that I mentioned the thin plates… 😉
Basically, if another program has done some registration on the data, it is usefull to know what kind of registration was used. A projective registration ( homographies ) only would be no problem. Otherwise it could be a problem.March 31, 2017 at 01:02 #13563
That did help in some areas, and made it worse in others. If I blend the two versions together, there is only one area where there is an issue. So Its getting very close to an entire flat, gradient correct, brightness single linear layer, cool.
I can give you some frames about the area where both struggled to blend, but as this is my next big mosaic project I wont be handing out all 25 frames to test. Sorry but they are years of data of my work so I like to keep it in house so to speak.
TomMarch 31, 2017 at 10:21 #13575
Hi Tom, no problem 😉
I totally understand, I am just really happy that you almost have it like you want 😉
Can you give an indication on how long this would have taken you if you tried it with your old mosaic routine using Registar, PI and PS?
I’ll look into the mosaic algorithms one of these days, to see if I can improve a bit more.
It would help me as well if, at a next mosaic run, you can give me a screen print of the registration RMS of all frames.
By the way, you are deselecting same camera and optics I presume? I did forget to mention that setting in my last message.
Since you are working with mono data with tight stars I presum,
Maybe try these settings if you haven’t done so already:
3) Analyse stars
minimum star size 4
clip star profile 0.1
enable filter star profile
And try to find as much stars as we can per panel
limit stars to 10.000
at 4) Register
firt try quadrilaterals
enable distortion correction
disable same camera and optics.
At a next run, try the triangles.
Let me know what happens 😉
MabulaApril 3, 2017 at 23:17 #13751
Previously it took me about 2 -3 weeks to create the base layer, align all files to this base layer, process them to non linear data then stitch and blend in PS.
Now if I can get the settings right, technically only 3 hours Laptop time. This would be fantastic, and would also leave me with linear data. However with experimentation, and the possibility of having to create separate mosaics to cover star mis alignment, then APP (plus some registar and PS) would take about 2 days to get to the same point.
I was unticking the Same camera setting. I’ll keep trying your suggestions for the other settings.
I just ran another 6 pane mosaic of an area where the stars were mis aligned. It worked perfectly. I blended it in PS. I must see can I registar and blend it while in Linear mode.
What do you mean by the RMS screen shot of the registration? Do you mean just the settings I have choosen, or what the program is doing in the smaller windows that pop up while the algorithms are running?