Noise level - sigma units?

[Falco Peregrinus]

Hi all,

Documentation for the sigma parameter says "Noise level in the image". How that noise is measured, what values can it accept, and in which "units"? What does it mean sigma 5 or sigma 100?
I found on some webpage that we should achieve around 100 detected sources, but I guess it depends on the type of image one has.
How to know which sigma to set, and can it be somewhat estimated?

Thanks in advance.

[Dustin Lang]

Hi,

"sigma" refers to the standard deviation of a Gaussian distribution, which is often what we use to describe the noise in astronomical images.  If you don't set --sigma, it will be estimated from the image.  I would recommend not setting it, unless the automatic estimation is failing.  (Can happen for processed JPEG images.)  It is in whatever units your image is in (ie, 0 to 255 for JPEG images; ADU for FITS).  More likely would be setting the --nsigma value, which sets the detection threshold -- a peak has to be above nsigma * sigma to be accepted as a source.

Like you say, for Astrometry.net to work you probably need about 20 detected sources.  If you can *see* stars in your image that aren't detected by the code, then you can think about changing --nsigma, but some images really just don't have that many detected stars, and then Astrometry.net is unlikely to work for those images.

cheers,

dustin

--
You received this message because you are subscribed to the Google Groups "astrometry" group.
To unsubscribe from this group and stop receiving emails from it, send an email to astrometry+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/18df6fcd-9df8-42e2-899a-da77b0cc64e1n%40googlegroups.com.

[Falco Peregrinus]

Thank you very much for such a clear explanation.
I have more than enough stars in my image,astrometry.net matched ~235 of them, with the smallest index file, although I can see that even some artifacts are detected.
Is it correct to assume that more matched stars, even though with more stars we see more scattering because smaller index file contains fainter stars on average, are leading to better mapping of every/random pixel?


Thanks

[Dustin Lang]

Hi,

It's a good question whether more matched stars (fainter, noisier) yield a better solution.  If we computed the solution using weights on stars depending on their uncertainties, then I think we could ensure that the solution always improved when adding more stars, but currently Astrometry.net doesn't do that, so it would be possible for the solution to get worse, or better, depending on the distributions!  It would also depend on how regular or distorted the solution is -- if you want to map out a complicated distortion pattern (or one that changes a lot at the edges), you'd need more stars.

cheers,

dustin

To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/aeb7c27d-ce35-4de7-8db7-27de8c3587f4n%40googlegroups.com.