joshw |
02/03/2007 08:20AM (Read 9020 times)
|
|
|
Status: offline
Registered: 12/09/2005
Posts: 26
|
Hi all,I'm looking for a good image registration task. I've played with imalign and xregister and I've run into are two problems. First, neither is very good at finding the shifts when the initial images are not well aligned (100 pixels or more offsets) and the offsets are unknown. This is a problem because I have lots of images and taking the time to estimate initial offsets for all of them would be a painful process. Second, neither task accounts for image rotation.I've used WCS information for this sort of alignment in the past, but no WCS has been defined for these images and doing so for each one would again be too time consuming.I know this is a challenging task, so any suggestions would be appreciated.thanks,
Josh
|
|
|
|
fitz |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 09/30/2005
Posts: 4040
|
Hi Josh,These wouldn't be NEWFIRM commissioning images would they? In any case, in the absence of WCS information you could use something like XYXYMATCH to do a catalog match of the detections (e.g. from STARFIND) to find the transformations. This will account for rotations/flips and as long as there is sufficient overlap in the catalogs normally works pretty well. If you have a rough idea of the plate center and know the plate scale pretty well, something like the WCSFIXER (http://nvo-iraf.noao.edu/wcsfixer) might be able to derive plate solution for your images and WREG could then do a better job. See the examples in various help pages (i.e. do a REFERENCES search on 'match' or 'register'), I think there are examples that might lead you through the various steps.Cheers,
-Mike
|
|
|
|
joshw |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 12/09/2005
Posts: 26
|
Hi Mike,Yes, it is for NEWFIRM data. How did you guess? It's also for a few other data sets that I'm working on as well. The XYXYMATCH task looks like it'll do the trick.thanks!
Josh
|
|
|
|
joshw |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 12/09/2005
Posts: 26
|
I'm having a little trouble with XYXYMATCH. It behaves well with some image pairs, but some (those with the longer shifts) don't work. I tried using interactive mode (specifying 3 stars in each image) and the initial linear transformations looks reasonable (appropriate shifts and rotations), but the matched triangles transformation defaults to zero shift and rotation. This is using the "triangles" matching.Here is the relevant output from the task:[code:1:88121e9250]
Initial linear transformation
xref[tie] = -24.41745 + 0.9988094 * x[tie] + -0.0064513 * y[tie]
yref[tie] = -110. + 0. * x[tie] + 1. * y[tie]
dx: -24.42 dy: -110.00 xmag: 0.999 ymag: 1.000 xrot: 0.0 yrot: 0.4Matched triangles transformation
xref[tie] = 0. + 1. * x[tie] + 0. * y[tie]
yref[tie] = 0. + 0. * x[tie] + 1. * y[tie]
dx: 0.00 dy: 0.00 xmag: 1.000 ymag: 1.000 xrot: 0.0 yrot: 0.05 reference coordinates matched
[/code:1:88121e9250]Any thought as to why the final transformation would reset to no shift, no rotation? There are several hundred reference coordinates, so matching only 5 is clearly wrong. When I use the "tolerance" matching (which requires 3 input tie points set interactively), I get 200+ matches based on that rough linear transformation.-Josh
|
|
|
|
fitz |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 09/30/2005
Posts: 4040
|
Hi Josh,I'd have to see the parameters you used and the coordinate lists to say anything definite. If you know an approximate shift you can use the LINTRAN task to offset the coordinates before trying the task if that seems to work better. It may just be that there was a bogus solution found, common errors are to set the 'nmatch' parameter too high, or to assume the 'tolerance' should be as large as the expected shift -- either of these will lead to a bunch of false positives.I poked around on tan and nutmeg trying to figure out where you're working but couldn't find anything. You can just give me this instead of sending images and I can grab the files from there. Looks like another clear night, some of the images I've seen from last night cleaned up pretty well, nice job.Cheers,
-Mike
|
|
|
|
joshw |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 12/09/2005
Posts: 26
|
Hi Mike,I've been working on other images up until now, but I've run into a similar problem now with the NEWFIRM images. I don't see why a false match would give exactly zero for the shifts and rotations. Perhaps I have some parameter or something set wrong.You can take a look at the data on nutmeg in ~4meter/newfirm/processed/. To see the commands I used look at the end of the process.cl file. The STDOUT output of XYXYMATCH is in xyxymatch_out.txt.thanks!
Josh
|
|
|
|
joshw |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 12/09/2005
Posts: 26
|
Hi Mike,I'm still stuck on the XYXYMATCH problem. Very few stars are matched in any solution even though the number of input stars is large (100-200). I've looked at the images and the stars found by DAOFIND and they look reasonable. Not every star in the images is found and some are false detections, but it seems to me that there are plenty of points in common.I'm no longer working on nutmeg as I am no longer on the mountain. I have the files locally on my laptop, so if you want to take a look, let me know and I can find a way to send you files.cheers,
Josh
|
|
|
|
fitz |
02/03/2007 08:20AM
|
|
|
Status: offline
Registered: 09/30/2005
Posts: 4040
|
Hi Josh,I played with the data a bit and was able to get a solution using 73 stars simply by increasing the 'nmatch' parameter a bit, i.e.[code:1:6cdc5ebd94] xyxymatch Stars_BNKLpos1_BrG_002_im1.txt Stars_BNKLpos1_BrG_001_im1.txt Match_BNKLpos1_BrG_002_im1.txt 2 interactive- verb+ tol=2 nmatch=50Input: Stars_BNKLpos1_BrG_002_im1.txt Reference: Stars_BNKLpos1_BrG_001_im1.txt Number of tie points: 0
Initial linear transformation
xref[tie] = 0. + 1. * x[tie] + 0. * y[tie]
yref[tie] = 0. + 0. * x[tie] + 1. * y[tie]
dx: 0.00 dy: 0.00 xmag: 1.000 ymag: 1.000 xrot: 0.0 yrot: 0.0Matched triangles transformation
xref[tie] = -235.2849 + 1.00239 * x[tie] + 0.00149924 * y[tie]
yref[tie] = -8.077873 + 0.00188834 * x[tie] + 1.001015 * y[tie]
dx: -235.28 dy: -8.08 xmag: 1.002 ymag: 1.001 xrot: 0.1 yrot: 359.973 reference coordinates matched
[/code:1:6cdc5ebd94]In general the tuning of the parameters is data dependent but I agree that simply printing zeroes isn't much of a clue that there was no solution, especially when it says there were N matched reference stars. In any case, this shift looks about right given the data I used. I assume you've already seen the pretty-picture first-light mosaics?Cheers,
-Mike
|
|
|
|