generateAlardLuptonBasisList

lsst.ip.diffim.generateAlardLuptonBasisList(config, targetFwhmPix=None, referenceFwhmPix=None, basisDegGauss=None, basisSigmaGauss=None, metadata=None)

Generate an Alard-Lupton kernel basis list based upon the Config and the input FWHM of the science and template images.

Parameters:
config : lsst.ip.diffim.PsfMatchConfigAL

Configuration object for the Alard-Lupton algorithm.

targetFwhmPix : float, optional

Fwhm width (pixel) of the template exposure characteristic psf. This is the _target_ that will be matched to the science exposure.

referenceFwhmPix : float, optional

Fwhm width (pixel) of the science exposure characteristic psf.

basisDegGauss : list of int, optional

Polynomial degree of each Gaussian (sigma) basis. If None, defaults to config.alardDegGauss.

basisSigmaGauss : list of int, optional

Sigmas of each Gaussian basis. If None, defaults to config.alardSigGauss.

metadata : lsst.daf.base.PropertySet, optional

If specified, object to collect metadata fields about the kernel basis list.

Returns:
basisList : list of lsst.afw.math.kernel.FixedKernel

List of basis kernels. For each degree value n in config.basisDegGauss (n+2)(n+1)/2 kernels are generated and appended to the list in the order of the polynomial parameter number. See lsst.afw.math.polynomialFunction2D documentation for more details.

Raises:
RuntimeError
  • if config.kernelBasisSet is not equal to “alard-lupton”
ValueError
  • if config.kernelSize is even
  • if the number of Gaussians and the number of given sigma values are not equal or
  • if the number of Gaussians and the number of given polynomial degree values are not equal

Notes

The polynomial functions (f) are always evaluated in the -1.0, +1.0 range in both x, y directions, edge to edge, with f(0,0) evaluated at the kernel center pixel, f(-1.0,-1.0) at the kernel (0,0) pixel. They are not scaled by the sigmas of the Gaussians.

Base Gaussian widths (sigmas in pixels) of the kernels are determined as:
  • If not all fwhm parameters are provided or config.scaleByFwhm==False then basisSigmaGauss is used. If basisSigmaGauss is not provided, then config.alardSigGauss is used. In both cases, the length of sigmas must be equal to config.alardNGauss.
  • If targetFwhmPix<referenceFwhmPix (normal convolution): First sigma Sig_K is determined to satisfy: Sig_reference**2 = Sig_target**2 + Sig_K**2. If it’s larger than config.alardMinSig * config.alardGaussBeta, make it the second kernel. Else make it the smallest kernel, unless only 1 kernel is asked for.
  • If referenceFwhmPix < targetFwhmPix (deconvolution): Define the progression of Gaussians using a method to derive a deconvolution sum-of-Gaussians from it’s convolution counterpart. [1] Only use 3 since the algorithm assumes 3 components.

Metadata fields

ALBasisNGauss : int
The number of base Gaussians in the AL basis functions.
ALBasisDegGauss : list of int
Polynomial order of spatial modification of the base Gaussian functions.
ALBasisSigGauss : list of float
Sigmas in pixels of the base Gaussians.
ALKernelSize : int
Kernel stamp size is (ALKernelSize pix, ALKernelSize pix).
ALBasisMode : str, either of config, convolution, deconvolution
Indicates whether the config file values, the convolution or deconvolution algorithm was used to determine the base Gaussian sigmas and the kernel stamp size.

References

[1]Ulmer, W.: Inverse problem of linear combinations of Gaussian convolution kernels (deconvolution) and some applications to proton/photon dosimetry and image processing. http://iopscience.iop.org/0266-5611/26/8/085002 Equation 40