Repository navigation
PSF classes should have a "flux" attribute #254
Description
We currently offer one class as a PSF, the GaussianPSF. The fitted parameters are sigma, amplitude and (x,y) position. In the end, we don't care about the amplitude, only about the total flux (given by the integral and depending on amplitude and sigma).
I have a feeling that there must be an analytic expression for this, but I'm right now a little confused by the fact that the docs https://github.com/astropy/photutils/blob/master/photutils/psf.py#L210 and the code https://github.com/astropy/photutils/blob/master/photutils/psf.py#L261 use a different normalisation and neither of them is 2/sqrt(pi) as I expected it to be http://en.wikipedia.org/wiki/Error_function.
@larrybradley @adonath : You two did the most recent changes to code and docs...
Activity
hamogu commentedon Mar 4, 2015
Note that the value of
psf/psf.max()also depends onx_0, y_0. If the center of the PSF shifts, then the pixel boundaries will fall differently and thus the evaluated numbers in the PSF will change, which will causepsf.max()to be different. Is that intentional? It might be right, I just cannot wrap my head around it right now.adonath commentedon Mar 4, 2015
If I remember correctly, the parameter used to be the total integrated flux (that's how I implemented it). I just noticed that @larrybradley changed the definition of the parameter in #218 to the peak amplitude, maybe because I called it
amplitudein the docs? (Which was a misnomer...sorry) I think it should be changed back. @larrybradley What do you think?Concerning the normalization factor: The factor
2/sqrt(2)is already included in the error function of scipy. The defintion of the 2D Gaussian using the error function needs a normalization factor of1/4, because the error function is normalized to one by definition, but only gives the one sided integral. For the full integral you need an additional factor of1/2in the 1d case and a factor1/2**2 = 1/4in the 2d case. Does that make sense?larrybradley commentedon Mar 5, 2015
@adonath, yes I completely agree that it should be changed back (when I was doing #218 I mentioned that
fluxseemed better thanamplitude, but nonetheless I continued your PR before the 0.1 release).Changes to Gaussian PSF
Changes to Gaussian PSF