LatticeFit.h

Classes

LatticeFit -- Fit every line of pixels parallel to any axis in a Lattice. (full description)

class LatticeFit

Interface

Public Members

static uInt fitProfiles (Lattice<Float>& outImage, Vector<Float>& fittedParameters, LinearFit<Float>& fitter, const Lattice<Float>& inImage, uInt whichAxis, const Vector<Bool>& fitMask, Bool returnResiduals)

static uInt fitProfiles (MaskedLattice<Float>* pOutFit, MaskedLattice<Float>* pOutResid, MaskedLattice<Float>& in, Lattice<Float>* pSigma, LinearFit<Float>& fitter, uInt axis, Bool showProgress=False)

Description

Prerequisite

Synopsis

For every line in the lattice parallel to axis number whichAxis (often axis number 2, typically the frequency axis in a spectral line cube) independently fit the functions in fitter at the positions where fitMask is true.

Example

Suppose one wanted to subtract a linear polynomial from every spectrum (3d axis) in an image. One could do this as follows:

    Image<Float> myImage("myimage"); // Get the image
    uInt nchan = myImage.shape()(2); // 0 relative axis number
    // Set up the fitter
    Polynomial<AutoDiff<Float> > linear(1);    
    LinearFitSVD<Float> fitter;
    fitter.setFunction(linear);
    Vector<Float> fittedParameters,

    // Set up a mask indicating what channels we want to fit over. We want
    // to fit over all channels.
    Vector<Bool> fitMask(nchan); fitMask = True;

    // Do the fit. True means subtract the fit from the model. In this case,
    // We overwrite the input with the output.
    fitProfiles (myImage, fittedParameters,fitter, myImage, 2, fitMask, True);

Motivation

Baseline fitting/continuum subtraction are important functions. This function essentially implements the IMLIN algorithm.

To Do

Save the model parameters in an (optional) other lattice.
Use logging classes, rather than the raw GlishSysEventSource.
Allow per-pixel weights.
Allow non-linear as well as linear LSQ fits.

fitting functions

Member Description

static uInt fitProfiles (Lattice<Float>& outImage, Vector<Float>& fittedParameters, LinearFit<Float>& fitter, const Lattice<Float>& inImage, uInt whichAxis, const Vector<Bool>& fitMask, Bool returnResiduals)

Fit baseline to lattice. Presently the fit parameters, other than the last one(s) in fitter, are lost. If returnResiduals is True, return data-fit, otherwise return the fit. For baseline and continuum subtraction, returnResiduals would normally be True.

static uInt fitProfiles (MaskedLattice<Float>* pOutFit, MaskedLattice<Float>* pOutResid, MaskedLattice<Float>& in, Lattice<Float>* pSigma, LinearFit<Float>& fitter, uInt axis, Bool showProgress=False)

Fit baseline to MaskedLattice. Fit and residuals can be optionally written (leave pointers at zero to not write out these lattices) You can optionally specify a weights lattice (1.0 if not given).