Setting a new reference frame: based on star analysis results

Could you post some screenshots of what you're doing and seeing? If you manually select the low resolution image to be added to the stack, it will be incorporated with the rest of the data, so I'm not sure what exactly you're trying. 🙂

What I do is the following:

1. Start with a low resolution wide field image, with enough stars visible to guide the alignment procedure, add this "alignment image" to the lights
2. Enter all panels for the mosaic to the lights
3. Select normal registration mode, dynamic distortion correction, not same camera and optics, and generous values for scale start and stop
4. Select enough stars in analyse stars (5000 does the job for me, but that surely depends on the field, exposure and mosaic
5. Perform all steps up to normalise
6. Deselect the low resolution image
7. Perform the final integration

Below are images showing the result without and with inclusion of the "alignment image", with a partial mosaic and one panel that somehow was off in position during acquisition.

 It's just that the registration step went through the entire list of images to match them against each other, while I already know in advance that the best choice is the "alignment image" and have marked this as such. I don't have log file entries available any more, they scrolled off the console by now. It just that this step took a few hours I think to complete, while I had expected something on the order of 5-10 minutes. Checking the console showed that all permutations of alignments were checked anyways.

Strange, I tried to reproduce the effect, this time with additional panels I'd shot last night, and the registration is done in 3 to 4 minutes.

Screenshot confirming the settings indeed just request normal registration:

Registration failed, while it definitely succeeded earlier...

50 minutes to get this far, what's the difference with my earlier attempt? Now I'm confused.

Registration failed on some of the frames !

Possible solutions for a successfull registration are:

- increase the automatic #stars target in 3) ANALYSE STARS
- disable same camera and optics (if enabled)
- try different pattern recognition descriptors
- increase the scale stop level
  especially if you have data of different image scales or
  with little overlap between the frames
- enable flip descriptors in X/Y
  if you need to register data from different telescopes

Edit:

Setting the Scale Stop from 10 to 15 solves the problem, nice alignment in less than 5 minutes. 

Thank you for the elaborate posts on the issue! I now get your use of the lower resolution reference frame, pretty clever I have to say. 😉 I think it is still taking this long as APP does need to have all data on all stars from each frame to be able to continue processing. The reference will just tell APP the FOV basically, it still needs to do all the calibration, star analysis etc. for the mosaic. I would still advice to perform the mosaic with the mosaic algorithms though, what's the reason for choosing regular integration?

For a regular mosaic one will more or less be forced to use dynamic distortion correction. This gives the alignment algorithm a lot more freedom to get a better fit, but also freedom to introduce crazy distortion where too little alignment information is present (noisy star positions e.g.).

Here's an example of what I got when doing a regular mosaic, note the somewhat distorted panel on the upper left, due to the very limited overlap between the panels:

And below are the same data, now aligned with the help of the alignment image, it even includes a totally detached frame where the pointing somehow got messed up during acquisition:

Also, the distortion in the upper left panel is gone now. And registration of the panels is *much* faster (if I use the correct settings that is...). Alignment now is done with order N (=37 in the final mosaic) alignments, while the mosaic alignment is order N*(N-1)/2 alignments (666 for my 6 x 6 panel mosaic plus alignment image, lots of sixes there). It really makes a difference, a few minutes versus a few hours.

Hi @ralph,

So by distortion, do you mean the panel is not a rectangle anymore? That is to be expected as APP is then correcting for distortion present in your optics. This is one of the very nice features even as it allows for even more precise alignment of the mosaic. I think it also allows for better panel to panel normalization, all processes that require way more advanced calculations and hence it takes longer. You can also first make parts of the mosaic separately and then perform a final mosaic with those panels.

Agreed that APP is a master in removing distortion. However, in the example above I used the same optics, so I would expect the same distortion correction in all panels. And as you can see in the shape of the other panels, as well as the second picture with the help of the alignment image, the distortion of the optics themselves is minimal (as can be expected from the Ritchey Chretien telescope design used in this case).

But coming back to my original question, judging by the way APP behaves depending on the (sometimes limiting) settings I used, I take it that if the regular alignement fails, APP automatically falls back to mosaic alignment?

No, it shouldn't. You have to specifically tell APP to do a mosaic, from the log I see it's doing normal registration but seems to repeat this all the time and changing the reference frame. I've never done a mosaic with normal registration, so I'm not sure what is happening exactly, it might be trying too hard or so as it isn't expecting a mosaic.. maybe I could have a look at the data as well?

Sure, where and under which identifier would you like me to upload them?

Thank you! You can upload some to our server (<- click that link) with username/password: appuser

Please create a directory called "ralph-mosaic" and I'll have a look asap.

Thanks, Vincent

Uploading now, should be done in a few minutes, most of it is there already.

Update: all there now.

Ok, I'm creating the mosaic now. RMS is 0.99 with distortion correction. I did not use a reference, low resolution image and used the following settings;

- 4000 stars
- Dynamic correction, and same camera/optics switched off
- Advanced normalize mode
- 1st degree LNC, 3 iterations (for such a big mosaic this is a slooow bit)
- MBB 10%
- Scale start: 4, scale stop: 10

I noticed that you have a little bit of coma in the corners, this is actual optical distortion which might be the reason that the frames are corrected slightly to improve the RMS.

This was the result;

And then with the light pollution correction once, mainly focusing some boxes to those more troublesome corners and just a few spread out over the image, which is already pretty well corrected (so very good data!);

And another pass with the LP-tool, multiple passes are ok and not destructive. I increase the boxes in size and more across the mosaic as the troublesome area corrected a bit better and didn't need such attention anymore. It is tricky with this much nebulosity though so I apply maximal stretch to better judge where to place them after a first correction, it can also become more clear where to do the second or more afterwards;

Increasing the contrast a little and adding some intelligent sharpening (as in APP doing this carefully, bringing out the gas structure more);

The troublesome corners are difficult indeed, the cause of those really should be dealt with at the data stage I think. Which we're discussing in the flat topic. 😉 But all in all, I love the result and I see no borders across the mosaic itself. I think this might work better if I correct these troublesome panels with light pollution correction, before creating the mosaic.

What is your take on it Mabula @mabula-admin ?

ps. Sorry these are screenshots so maybe difficult to see really well.

Thanks for the verificationprocessing! 😉

Just curious: how long did the registration step take?

No problem, if you have any additional questions, just shoot.

I can't remember anymore the time of registration, I do have a rather new and fast computer so that was pretty fast. The LNC 3 took by far the most time, I'm guessing in total it took about an hour. I might be off there though, would have to time that. 🙂