Comments on: Fourier Transforms of Images in Python http://www.astrobetter.com/fourier-transforms-of-images-in-python/ Tips and Tricks for Professional Astronomers Mon, 06 Feb 2012 12:32:48 +0000 hourly 1 http://wordpress.org/?v=3.2.1 By: Adam Ginsburg http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-16442 Adam Ginsburg Mon, 08 Aug 2011 20:51:02 +0000 http://www.astrobetter.com/?p=1809#comment-16442 Omar: I looked into this, but apparently the process is to make a "scikit" first. I'll look into doing that, but not immediately. In the meantime, I've split up agpy into some sub-packages. You should now be able to access PSDs without any unnecessary dependencies from the module AG_fft_tools. Omar: I looked into this, but apparently the process is to make a “scikit” first. I’ll look into doing that, but not immediately.

In the meantime, I’ve split up agpy into some sub-packages. You should now be able to access PSDs without any unnecessary dependencies from the module AG_fft_tools.

]]> By: Omar http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-16440 Omar Mon, 08 Aug 2011 20:32:44 +0000 http://www.astrobetter.com/?p=1809#comment-16440 @Adam: the problem was, when i import agpy.psds python would try to parse agpy.__init__ and that includes all the dependencies. if i might wish me a pony: scipy.fftpack might actually need a psd function, so i would say this is something you might want to suggest to scipy for inclusion! @Adam: the problem was, when i import agpy.psds python would try to parse agpy.__init__ and that includes all the dependencies.

if i might wish me a pony: scipy.fftpack might actually need a psd function, so i would say this is something you might want to suggest to scipy for inclusion!

]]> By: Adam Ginsburg http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-16436 Adam Ginsburg Mon, 08 Aug 2011 20:18:07 +0000 http://www.astrobetter.com/?p=1809#comment-16436 Omar - Yes, the agpy package has a lot of extra requirements... it's not really packaged for distribution at the moment. However, the PSD and radialprofile codes should stand on their own. Would it be helpful if I packaged them separately? If so, I'll work on that. One thing to note - the radial profile code included in the above post, and that I had previously used, is very slow for large images. There are some nifty tricks using numpy.histogram to make it faster. I've implemented them here: http://code.google.com/p/agpy/source/browse/trunk/agpy/radialprofile.py but in short, the fast way to make the radial profile, once you've made a "radius" array and determined the bin locatinos, is: np.histogram(r, bins, weights=image) Omar – Yes, the agpy package has a lot of extra requirements… it’s not really packaged for distribution at the moment. However, the PSD and radialprofile codes should stand on their own. Would it be helpful if I packaged them separately? If so, I’ll work on that.

One thing to note – the radial profile code included in the above post, and that I had previously used, is very slow for large images. There are some nifty tricks using numpy.histogram to make it faster. I’ve implemented them here:
http://code.google.com/p/agpy/source/browse/trunk/agpy/radialprofile.py

but in short, the fast way to make the radial profile, once you’ve made a “radius” array and determined the bin locatinos, is:
np.histogram(r, bins, weights=image)

]]> By: Omar http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-16435 Omar Mon, 08 Aug 2011 19:51:46 +0000 http://www.astrobetter.com/?p=1809#comment-16435 Thank you very much. In radialProfile.py line 17 instead of center = np.array([(x.max()-x.min())/2.0, (x.max()-x.min())/2.0]) should read center = np.array([(x.max()-x.min())/2.0, (y.max()-y.min())/2.0]) in my opinion! @Adam: i did try your code prior to this one, but it has many more dependencies (like pyfits), which aren't really necessary for psd, imho. Thank you very much. In radialProfile.py line 17 instead of

center = np.array([(x.max()-x.min())/2.0, (x.max()-x.min())/2.0])

should read

center = np.array([(x.max()-x.min())/2.0, (y.max()-y.min())/2.0])

in my opinion!
@Adam: i did try your code prior to this one, but it has many more dependencies (like pyfits), which aren’t really necessary for psd, imho.

]]> By: Adam Ginsburg http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-2193 Adam Ginsburg Wed, 24 Mar 2010 03:30:23 +0000 http://www.astrobetter.com/?p=1809#comment-2193 The tricks you used in radialProfile.py are much faster than the not-so-clever tricks I used in my PSD code. However, it excluded inner & outer points, and I wanted the standard deviation too.... so modified code is posted at <a href="http://www.astrobetter.com/wiki/tiki-index.php?page=python_radial_profiles" rel="nofollow"> the wiki</a> and <a href="http://code.google.com/p/agpy/source/browse/trunk/agpy/radialprofile.py" rel="nofollow">agpy</a>. The tricks you used in radialProfile.py are much faster than the not-so-clever tricks I used in my PSD code. However, it excluded inner & outer points, and I wanted the standard deviation too…. so modified code is posted at the wiki and agpy.

]]> By: Ian J. Crossfield http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-2089 Ian J. Crossfield Thu, 11 Mar 2010 05:52:50 +0000 http://www.astrobetter.com/?p=1809#comment-2089 A few years back I wrote a routine similar to your "radialProfile" that computes average, mean, min/max in annuli of an image, with the option to define the image coordinate system (and thus the center-of-interest). If anyone wants to take a look, I've put it online <a href="http://www.astro.ucla.edu/~ianc/files/radial_data.py" rel="nofollow">here</a>. For a centered image, the syntax is just: <code> import pylab as py from radial_data import radial_data data = radial_data(image, annulus_width=2, rmax=100) py.plot(data.r, data.mean) </code> A few years back I wrote a routine similar to your “radialProfile” that computes average, mean, min/max in annuli of an image, with the option to define the image coordinate system (and thus the center-of-interest). If anyone wants to take a look, I’ve put it online here.

For a centered image, the syntax is just:

import pylab as py
from radial_data import radial_data
data = radial_data(image, annulus_width=2, rmax=100)
py.plot(data.r, data.mean)

]]> By: Adam Ginsburg http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-2030 Adam Ginsburg Thu, 04 Mar 2010 20:48:19 +0000 http://www.astrobetter.com/?p=1809#comment-2030 Sure thing. I have test code + a test image: http://code.google.com/p/agpy/source/browse/trunk/tests/ http://code.google.com/p/agpy/source/browse/trunk/tests/example_psd.py http://code.google.com/p/agpy/source/browse/trunk/tests/PSF.fits The example shows the power spectrum of a non-Gaussian PSF with sidelobes that should approximately correspond to Airy rings. I don't bother with the 2d PSD (even though the code is written for that) because I don't really use PSDs. Ivan: The autocorrelation (rather than the PSD) should contain information about the psf, I think. Though when I tried it myself it looks like the autocorrelation^2 has ~the same FWHM... I'm not sure exactly why that should be. Sure thing. I have test code + a test image:
http://code.google.com/p/agpy/source/browse/trunk/tests/
http://code.google.com/p/agpy/source/browse/trunk/tests/example_psd.py
http://code.google.com/p/agpy/source/browse/trunk/tests/PSF.fits

The example shows the power spectrum of a non-Gaussian PSF with sidelobes that should approximately correspond to Airy rings. I don’t bother with the 2d PSD (even though the code is written for that) because I don’t really use PSDs.

Ivan: The autocorrelation (rather than the PSD) should contain information about the psf, I think. Though when I tried it myself it looks like the autocorrelation^2 has ~the same FWHM… I’m not sure exactly why that should be.

]]> By: Ivan Zolotukhin http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-2029 Ivan Zolotukhin Thu, 04 Mar 2010 17:35:47 +0000 http://www.astrobetter.com/?p=1809#comment-2029 Roy, this is just a background noise causes the constant term, check e.g. eq. 6-7 in Sheehy et al. 2006. Jessica, thanks for your comments. I'm after another problem actually: very crowded field photometry. Seems that I have no chance to estimate PSF from the image itself, so I'm exploring what FT can give here... Roy, this is just a background noise causes the constant term, check e.g. eq. 6-7 in Sheehy et al. 2006.

Jessica, thanks for your comments. I’m after another problem actually: very crowded field photometry. Seems that I have no chance to estimate PSF from the image itself, so I’m exploring what FT can give here…

]]> By: Jessica http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-2028 Jessica Thu, 04 Mar 2010 17:05:42 +0000 http://www.astrobetter.com/?p=1809#comment-2028 This is an adaptive optics image, so we have approximately a diffraction limited core + seeing halo. So at those intermediate frequencies, you are seeing the AO correction. This is an adaptive optics image, so we have approximately a diffraction limited core + seeing halo. So at those intermediate frequencies, you are seeing the AO correction.

]]> By: Jessica http://www.astrobetter.com/fourier-transforms-of-images-in-python/#comment-2027 Jessica Thu, 04 Mar 2010 16:59:59 +0000 http://www.astrobetter.com/?p=1809#comment-2027 Adam, you hit the nail on the head. I know you can use fftfreq to construct the spatial frequencies, but I just haven't gotten around to constructing the proper 1D X and Y (or nu_x and nu_y) axes or a 2D meshgrid. Thanks for the example! Would you be interested in sharing your code (or a link to it) on the wiki and if so, can you give me a brief example of how you use it? Ivan, right now I am applying the Sheehy et al. method to derive an aperture correction for crowded field adaptive optics photometery (1D PSF estimation from image + known positions of stars + model of atmosphere/AO system/telescope). Currently their code is in IDL, so I have been using python to call out to idl (via pIDLy). I wanted to explore whether it would be worth porting the code to python; especially since I was contemplating whether the code could be expanded to 2D. However, I think a more productive direction would be the methods/code in the papers by Matthew Britton (use empirical PSF from somewhere in an image + Cn2 profile of atmosphere to reconstruct the PSF at all other points in the image). I am using laser guide star AO with an off-axis tip-tilt star so this complicates things enormously. Adam, you hit the nail on the head. I know you can use fftfreq to construct the spatial frequencies, but I just haven’t gotten around to constructing the proper 1D X and Y (or nu_x and nu_y) axes or a 2D meshgrid. Thanks for the example! Would you be interested in sharing your code (or a link to it) on the wiki and if so, can you give me a brief example of how you use it?

Ivan, right now I am applying the Sheehy et al. method to derive an aperture correction for crowded field adaptive optics photometery (1D PSF estimation from image + known positions of stars + model of atmosphere/AO system/telescope). Currently their code is in IDL, so I have been using python to call out to idl (via pIDLy). I wanted to explore whether it would be worth porting the code to python; especially since I was contemplating whether the code could be expanded to 2D. However, I think a more productive direction would be the methods/code in the papers by Matthew Britton (use empirical PSF from somewhere in an image + Cn2 profile of atmosphere to reconstruct the PSF at all other points in the image). I am using laser guide star AO with an off-axis tip-tilt star so this complicates things enormously.

]]>