AIDA for lunar photometry

[Ryan Anderson]

Hello,

It looks like this discussion group is not very active, but I thought I would try posting here anyway. I am interested in using AIDA on a set of ground-based observations of the moon. The intent of the project is to collect accurate photometric measurements of the lunar disk, even when atmospheric scattering or sub-optimal focus spreads the signal out beyond the extent of the lunar disk. My hope is that deconvolving the images will correct for this and collect the lunar signal back where it should be. Each lunar observation is accompanied by observations of point sources (stars) from the same night that can be provided to AIDA as initial guesses at the point-spread function, however the star images are half the size of the lunar images (128x128 rather than 256x256). The individual image files are multispectral, so they contain a stack of images of the same object (the moon or a star) observed with multiple narrow-band filters (e.g. 256x256x9).

So, with that background, my questions are:

1. Does it sound like AIDA would be suitable for the goal of photometrically accurate deconvolution of lunar images? Do you know of published results using AIDA for photometric applications?

2. Can AIDA deal with PSF images that are not the same size as the object images? If not, how would you recommend modifying the smaller PSF to work? Just center them inside a blank image of the correct size?

3. How close to the true PSF do the user-supplied PSF files need to be? (I ask in case the observing conditions changed significantly over the course of the night) How many PSF images are recommended?

3. Can AIDA deal with multi-band images, or should the bands be deconvolved as individual images?

Thanks!

[Erik F. Y. Hom]

Dear Ryan,

Thanks for your post; yes, the discussion group is not very active, but hopefully that will change as I get help to build out AIDA. Per your questions, answers are embedded below:

On May 20, 2014, at 7:30 PM, Ryan Anderson wrote:

> Hello,
> It looks like this discussion group is not very active, but I thought I would try posting here anyway. I am interested in using AIDA on a set of ground-based observations of the moon. The intent of the project is to collect accurate photometric measurements of the lunar disk, even when atmospheric scattering or sub-optimal focus spreads the signal out beyond the extent of the lunar disk. My hope is that deconvolving the images will correct for this and collect the lunar signal back where it should be. Each lunar observation is accompanied by observations of point sources (stars) from the same night that can be provided to AIDA as initial guesses at the point-spread function, however the star images are half the size of the lunar images (128x128 rather than 256x256). The individual image files are multispectral, so they contain a stack of images of the same object (the moon or a star) observed with multiple narrow-band filters (e.g. 256x256x9).
>
> So, with that background, my questions are:
> 1. Does it sound like AIDA would be suitable for the goal of photometrically accurate deconvolution of lunar images? Do you know of published results using AIDA for photometric applications?

Yes--the original AIDA paper actually describes some photometric results, and Franck Marchis has routinely used AIDA for photometric reconstructions. I would ask him, but you could also check his papers (e.g., on asteroids).

> 2. Can AIDA deal with PSF images that are not the same size as the object images? If not, how would you recommend modifying the smaller PSF to work? Just center them inside a blank image of the correct size?

Yes, this should work without you needing to do anything, i.e., the PSFs should just be rescaled to be bigger. You can also just center them inside a blank image, but I recommend you do this properly (not ideal at the moment, but correct for the image processing pipeline): add it to a back ground "blank" that has representative pixel values of the noise in your system.

> 3. How close to the true PSF do the user-supplied PSF files need to be? (I ask in case the observing conditions changed significantly over the course of the night) How many PSF images are recommended?

I suggest you get the true PSFs as close to reality as possible. And the more (truly representative) PSFs the better--classic stats issue. There is no magic and the information you are using to constrain the decon is very important for good results. A good estimate of the true PSF is one of the more important things even with AIDA.

> 3. Can AIDA deal with multi-band images, or should the bands be deconvolved as individual images?

AIDA can't process multi-band image cubes properly yet--though this is just a "structural"/data frame processing issue, not a fundamental limit of the underlying algorithm. For what you are doing, separate out the bands as individual 2D images and then decon.

Hope this helps! Good luck and keep us posted if you have problems (or good results!).

Best,
Erik

[Ryan Anderson]

Hi Erik,

Thanks for your response!

I tried running AIDA on a test case: a sharp-edged bright circle in an otherwise blank image, convolved with a 2-D gaussian (see attached testimage.jpg and testpsf.jpg). When I run AIDA on this, providing the true PSF and the blurred test image as inputs, I get an unusual-looking PSF as output (see attached) that looks as if the true psf has been split into 4 pieces. Any idea why this might be?

I also tried using some of our actual data, and the deconvolution seems to mostly work, but the PSF being smaller than the main image seems to cause the output to be split into pieces (see attached - image with "mm230451" in the name). Is there a parameter that I need to adjust to deal with PSF images that are not the same size as the object images, or does this indicate that I should just try to match the sizes?

Thanks,

Ryan

[Erik F. Y. Hom]

Hi Ryan,

> I tried running AIDA on a test case: a sharp-edged bright circle in an otherwise blank image, convolved with a 2-D gaussian (see attached testimage.jpg and testpsf.jpg). When I run AIDA on this, providing the true PSF and the blurred test image as inputs, I get an unusual-looking PSF as output (see attached) that looks as if the true psf has been split into 4 pieces. Any idea why this might be?
>
> I also tried using some of our actual data, and the deconvolution seems to mostly work, but the PSF being smaller than the main image seems to cause the output to be split into pieces (see attached - image with "mm230451" in the name). Is there a parameter that I need to adjust to deal with PSF images that are not the same size as the object images, or does this indicate that I should just try to match the sizes?

This must be a silly bug in the resizing protocol. I'll need to fix this. Could you email me separately (er...@fyhom.com) with the files you used to generate the splitting?

For the immediate goal of getting results, I suggest you embed your PSFs into a background image of 256x256 detector noise as suggested previously. Then run again with equi-sized image array inputs. Splitting hopefully should go away...if not, will need to dig and fix this...

Let me know what you get!

Best,
Erik