Image Preprocessing Techniques for Astrometry in Adverse Weather

[Sharaf Mohanna]

If the weather was not clear when I took the photo, can astrometry still work on the image?
If the photo contains clouds, how can this be handled in order to estimate the right ascension (RA) and declination (Dec)?
Are there any algorithms that can preprocess the image to reduce blur, noise, or fog, so that astrometric calibration can still be performed?  

[Dustin Lang]

Hi,

For astrometry.net, you usually need 12-15 of the brightest stars to be detected.

Fundamentally, detecting stars is all about signal-to-noise.  In an idealize CCD image, you have "sky" background, which varies pixel-to-pixel, giving you noise, as well as readout noise.  And then the signal from the stars gets spread out into multiple pixels by the Point Spread Function.  Uniform clouds would just reduce the amount of signal.

In the web version, if you have an image with a wide PSF (blurry), you can use the "Downsample" option (default = 2).  With the "solve-field" command, you can use "--downsample"; there's also the "--nsigma" option that says how much signal-to-noise is needed to detect a star - good when there are clouds or noise making the stars faint.

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 visit https://groups.google.com/d/msgid/astrometry/3fcb0ece-aa3d-4e84-8c23-2ec238d082b0n%40googlegroups.com.

[Sharaf Mohanna]

Hi Dustin,

I have a question regarding my simulation results.

When I simulate partially cloudy sky conditions (sample attached), I obtain good results from astrometry.net. The solution matches both the simulation ground truth and my algorithm.

However, when I simulate fully cloudy weather (sample attached) and add noise to the star field, the results from astrometry.net degrade significantly. In many cases, I do not obtain a valid solution from the website, and my algorithm also fails to produce any output.

These are the parameters I am currently using:
SOLVE_ARGS = [
    "--overwrite",
    "--dir", SOLVED_DIR,
    "--downsample", "2",
    "--scale-units", "degwidth",
    "--scale-low", "75",
    "--scale-high", "110",
    "--config", BACKEND_CFG,
    "--crpix-center",
    "-v",
    "--continue",
    "--no-plots",
    "--cpulimit", "180",
]

Do you think astrometry.net is fundamentally unable to solve images under heavy or full cloud conditions due to insufficient detectable stars?

If not, are there specific parameters you would recommend adjusting to improve robustness for images like the attached full-cloud example in algrithm and website?

Additionally, I am not certain whether my noise model under fully cloudy conditions is physically realistic. It may introduce excessive high-frequency texture or unrealistic contrast. Could this be contributing to the solver’s failure? If so, do you have recommendations on how to model clouds and noise more realistically for astrometric solving?

Finally, could you point me to documentation or references explaining how astrometry.net performs star detection and how noise and thresholding affect the solving process?

Thank you,

[Dustin Lang]

Hi,

It is easy to see what is going wrong with your "fully clouded" example - on the web site, click on the "source extraction image" link -- eg, attached.  Some of the blobs of illuminated cloud are getting detected as "stars", and then of course astrometry.net cannot work.

Since your "cloud" simulation looks like it is generating power at certain scales, it would probably be possible to filter out those scales.  The very simple Astrometry.net star-detection algorithms will not do that.  We do estimate and remove a slowly-varying background level.

The paper describes this, I believe.

https://ui.adsabs.harvard.edu/abs/2010AJ....139.1782L/abstract

cheers,

dustin

To view this discussion visit https://groups.google.com/d/msgid/astrometry/2f0e288c-699c-485e-9a93-3bad87979f0fn%40googlegroups.com.