Combining Images from 2 Different cameras

[pfile]

to be honest i think that the problem here is that there's something wrong with the calibration of the images. if you zoom in on those subs you'll see that there are black specks all over the place in both images. seems worse in the 16803 image. in the 16200 image there are a lot of what appear to be hot pixels, and what's happening is that the star detector is thinking that many of those hot pixels are stars. faced with loads of bogus stars, SA can't find a match. however, when you define a preview like mike has done, that limits the # stars sent to the RANSAC stage and it does find a match. with images this similar there actually should be no need to define previews so i feel the actual problem is the star detection phase.

so i wonder if the calibration went wrong somehow, either possibly with the darks not matching or perhaps scaled too much.

anyway, i was able to register the 16803 image to the 16200 image by first mirroring the 16803 image as mike did, then without any previews, running SA after increasing Hot Pixel Removal to 2 and Noise Reduction to 2. it actually finds a solution with only hot pixel removal set to 2, but tries a lot harder. by increasing the noise reduction it solves right away.

anyway the flow for debugging problems like these in general is to set SA to "detected stars" mode and see what you get. if there are a bunch of crosshairs on things that aren't stars, it's a GIGO situation. you can try tweaking the star detection parameters to improve the detection quality and go from there.

rob

[jhayes_tucson]

Mike,
Wow!  That's exactly what I'm trying to do, except that I can't get it to work.  I think that I did exactly what you did and I get the errors shown on the first screen shot.  The second screen shot shows the previews that I picked.  I'm loading the 16200 view as the reference and I've tried aligning the preview and the full image and neither works.  What other settings did you use??  I'm either missing something or something is screwy.

John

[jhayes_tucson]

to be honest i think that the problem here is that there's something wrong with the calibration of the images. if you zoom in on those subs you'll see that there are black specks all over the place in both images. seems worse in the 16803 image. in the 16200 image there are a lot of what appear to be hot pixels, and what's happening is that the star detector is thinking that many of those hot pixels are stars. faced with loads of bogus stars, SA can't find a match. however, when you define a preview like mike has done, that limits the # stars sent to the RANSAC stage and it does find a match. with images this similar there actually should be no need to define previews so i feel the actual problem is the star detection phase.

so i wonder if the calibration went wrong somehow, either possibly with the darks not matching or perhaps scaled too much.

anyway, i was able to register the 16803 image to the 16200 image by first mirroring the 16803 image as mike did, then without any previews, running SA after increasing Hot Pixel Removal to 2 and Noise Reduction to 2. it actually finds a solution with only hot pixel removal set to 2, but tries a lot harder. by increasing the noise reduction it solves right away.

anyway the flow for debugging problems like these in general is to set SA to "detected stars" mode and see what you get. if there are a bunch of crosshairs on things that aren't stars, it's a GIGO situation. you can try tweaking the star detection parameters to improve the detection quality and go from there.

rob

Rob,
Yeah, I'm not sure what's up with the black specks in the 16803 data.  I'll have to go back to add a pedestal to make sure that nothing is getting clipped when that data is calibrated.  I snagged calibrated data from 3 years ago so it may be a challenge to fix that.  When I first tried this, I spent some time refining the star selection parameters and I see crosses on 99% of the stars.  As you can see I set the hot pixel and NR to 2 as well with no success.  The calibrated 16200 data looks 4 orders of magnitude better than the raw data so I assume that any remaining hot pixels are due to uncorrected hot pixels or cosmic rays.  Other than adding a pedestal to fix any clipped black pixels, I'm not sure what I can do about any residual bright pixels.  Clearly, you guys can get it to work so I'm doing something wrong here...

John

[pfile]

i had in the star detection pane, from top to bottom:

5
0
2
2
-1.0
0.8
0.5
1.0

these are the defaults except for the hot pixel removal and noise reduction.

with those settings i see 1207 stars in the 16200 image and 2932 stars in the 16803 image. with everything set to defaults, there were 1817 stars detected in the 16200 image and 4395 stars found in the 16083 image. most of those 4395 stars are bogus of course and i think that's the root of the problem.

rob

[msmythers]

John

I had used defaults but I decided to do the process one more time Reset StarAlignment to defaults and repeated the process with the same results.


Mike

[jhayes_tucson]

Rob,
Let me look at that.  Here's what I see with my settings (8,3,2,3,3,0.3,.98,0.5,1.0).  [That's just my current setting...I've tried quite a few at this point.] It is finding 101 and 80 stars in the two regions with 46 matches.  It finds the angular alignment and then it falls apart.  I checked and with the settings that I'm using it is seeing only stars.  The screen shot shows what it finds.  I have something set wrong because it is ignoring a lot of the stars so I'll have to figure out why.  I'll try your settings to see what it does next...

John

[jhayes_tucson]

Rob,
I tried your settings and it finds a lot more stars.  I checked and again it looks like it is picking up real stars (many are super faint.)  Here's what it reports before it fails.  446 and 388 stars in each region.  That should be enough...yet it fails.
John

[pfile]

are you using the same two images that are on dropbox? that's the only thing i can think of that might cause this. or - are your RANSAC settings changed from the default?

rob

[jhayes_tucson]

Rob,
Yes, I'm using the same images that I put in Drop Box and I'm using the default RANSAC settings.  I don't think that it should make any difference but I'm running on a MacBook Pro and I've checked that everything is up-to-date.  This is very strange.  I'm going to go off to mess with re-booting everything and fooling with it some more.  I agree that both you and Mike have made it work so it's not Voodoo.  Something is screwy.

Thanks for all your help and advice!
John

[pfile]

here's one more data point:

i applied CC to both images with auto detect - hot sigma 2.1 and cold sigma 1.4 (kind of aggresive i guess but only 1000s of pixels were picked up), then opened the two new images. i flipped the 16803 image horizontally, reset the SA interface and then aligned the 16803 image to the 16200 image without changing anything in SA. it took 15 tries but it found the solution...

rob

[jhayes_tucson]

Rob,
Ah...thanks, that's good to know.  I'll try that to see if I can get it to work.  Ultimately, I think that I'll have to revisit the image calibration step.  I agree that it needs to be cleaner.  I generally let dithering and the stacking filter clean up this stuff but that's clearly not going to work in this case.
John

[msmythers]

I just ran this without any previews and it solved on the 15th try. Sorry for the low font dpi but I wanted to show the whole process in the process console. Open files, mirror horizontal and just defaults for SA.


Mike

[jhayes_tucson]

Rob,
Bingo!  I increased the RANSAC tolerance from 0.0 to 0.6 and it worked first try.  Now I see that you have the same parameter set to 2.00.  That appears to have been the problem.  Thanks a million for all of your help!    A++.

John

[pfile]

so that's what i was saying - 2.0 is the default ransac tolerance. it seems you must have some saved process icons that you're loading SA from?

something that's weird in all of this is that i swore that SA could handle mirrored images as well, but it appears that's not the case, at least here.

rob

[msmythers]

rob

I've had some images work when they are flipped or other cardinal rotations and some that don't. I think the noise and or hot pixels are the probable problem. I just got into the practice of doing a fast rotation(no interpolation) for all images effected in a set to make life easier especially since I like to use Blink before doing any calibration.


Mike