ConvolutionEquation.h

Classes

ConvolutionEquation -- Implements the convolution equation (full description)

class ConvolutionEquation: public LinearEquation< Array<Float>, Array<Float> >

Interface

Public Members
ConvolutionEquation()
ConvolutionEquation(const Array<Float> & psf, const Array<Float> & dirtyImage)
ConvolutionEquation(const Array<Float> & psf, const MaskedArray<Float> & dirtyImage)
~ConvolutionEquation()
virtual Bool evaluate(Array<Float> & result, const LinearModel< Array<Float> > & model)
Bool evaluate(Array<Float> & result, const LinearModel< MaskedArray<Float> > & model)
Bool evaluate(MaskedArray<Float> & result, const LinearModel< MaskedArray<Float> > & model)
Bool evaluate(Array<Float> & result, const IPosition & position, const Float amplitude, const IPosition & modelShape)
virtual Bool residual(Array<Float> & result, const LinearModel< Array<Float> > & model)
virtual Bool residual( Array<Float> & result, Float & chisq, const LinearModel< Array<Float> > & model)
virtual Bool residual( Array<Float> & result, Float & chisq, Array<Float> & mask, const LinearModel< Array<Float> > & model)
Bool residual(Array<Float> & result, const LinearModel< MaskedArray<Float> > & model)
Bool residual(MaskedArray<Float> & result, const LinearModel< MaskedArray<Float> > & model)
IPosition psfSize()
void flushPsf()

Description

Review Status

Date Reviewed:
yyyy/mm/dd

Prerequisite

Etymology

This class implements convolution within the LinearEquation framework.

Synopsis

This class is used in conjunction with classes like HogbomCleanModel to implement deconvolution algorithms. This class contains the point spread function (psf) and the convolved data (dirty image), and is able to convolve a supplied model with the psf to produce a predicted output (using the evaluate() function), or to subtract the convolved data and produce a residual (using the residual() function).

See the documentation for HogbomCleanModel for an example of how this class can be used to perform deconvolution.

This class also contains specialised functions (like the version of evaluate() for a point source model) that speed up the calculation of the convolution. This specialised version of evaluate() does not need to actually perform the convolution and instead returns a suitable part of the psf (zero padded if necessary). When this function is called this class will get the psf from the convolver and cache it, on the assumption that many evaluations of this function will be requested (as occurs in Clean algorithms).

The size and shape of the psf and the supplied model may be different. The only restriction is that the dimension of the psf must be less than or equal to the dimension of the model. If the dimension of the model is larger than the dimension of the psf then the convolution will be repeated along the slowest moving (last) axis. The dirty image and the supplied model must be the same size and shape.

This class can also operate on MaskedArrays (and models representable by MaskedArrays). But the mask is currently discarded and the convolution performed on the entire supplied model. This may change in the future.

Example

    Matrix<Float> psf(4,4), dirty(20,20), model(20,20);
    .... put some meaningful values into these Arrays....
    // create a convolution equation, and an array model
    ConvolutionEquation convEqn(psf, dirty);
    ArrayModel<Float> myModel(model);
    // now calculate the convolution of the model and the psf
    Matrix<Float> prediction;
    convEqn.evaluate(myModel, prediction);
    // and calculate the difference between the predicted and actual convolution
    Matrix<Float> residual;
    convEqn.residual(mymodel, residual)
    

Motivation

This class was designed with deconvolution in mind.

To Do

Member Description

ConvolutionEquation()

Construct the ConvolutionEquation. Until I write some functions for setting the private data the default constructor is essentially useless

ConvolutionEquation(const Array<Float> & psf, const Array<Float> & dirtyImage)

Construct the ConvolutionEquation setting the psf and measured data

ConvolutionEquation(const Array<Float> & psf, const MaskedArray<Float> & dirtyImage)

Construct the ConvolutionEquation setting the psf and measured data Even though a MaskedArray is used as an arguement the mask is discarded internally and hence not used by residual().

~ConvolutionEquation()

Somewhere I read that a destructor should alway be defined even if it does nothing (as this one does).

virtual Bool evaluate(Array<Float> & result, const LinearModel< Array<Float> > & model)

Do the convolution of the model supplied by the LinearModel class with the internal psf. Return the answer in result .

Bool evaluate(Array<Float> & result, const LinearModel< MaskedArray<Float> > & model)

Do the convolution of the model supplied by the LinearModel class with the internal psf. Return the answer in result. This version uses Masked arrays. but the mask is currently discarded internally.

Bool evaluate(MaskedArray<Float> & result, const LinearModel< MaskedArray<Float> > & model)

Do the convolution of the model supplied by the LinearModel class with the internal psf. Return the answer in result. This version uses MaskedArrays, but the mask is not currently used. However the model mask is transfered to the result unchanged.

Bool evaluate(Array<Float> & result, const IPosition & position, const Float amplitude, const IPosition & modelShape)

Do the convolution of the a point source model at position 'position' with amplitude 'amplitude' and the internal psf. Return the answer in result.

virtual Bool residual(Array<Float> & result, const LinearModel< Array<Float> > & model)

Calculate the convolution of the model (supplied by the LinearModel class) and the psf and the difference between this and the supplied (presumably measured) convolution.

virtual Bool residual( Array<Float> & result, Float & chisq, const LinearModel< Array<Float> > & model)

Calculate the convolution of the model (supplied by the LinearModel class) and the psf and the difference between this and the supplied (presumably measured) convolution. Also return chisq.

virtual Bool residual( Array<Float> & result, Float & chisq, Array<Float> & mask, const LinearModel< Array<Float> > & model)

Calculate the convolution of the model (supplied by the LinearModel class) and the psf and the difference between this and the supplied (presumably measured) convolution. Also return chisq, considering a mask image

Bool residual(Array<Float> & result, const LinearModel< MaskedArray<Float> > & model)

Calculate the convolution of the model (supplied by the LinearModel class) and the psf and the difference between this and the supplied (presumably measured) convolution. This version uses Masked arrays. but the mask is currently discarded internally.

Bool residual(MaskedArray<Float> & result, const LinearModel< MaskedArray<Float> > & model)

Calculate the convolution of the model (supplied by the LinearModel class) and the psf and the difference between this and the supplied (presumably measured) convolution. This version uses Masked arrays. but the mask is currently discarded in the calculations and transfered unchanged from the model to the result.

IPosition psfSize()

return the psf size used in the convolution. The returned size does not include any zero padding

void flushPsf()

release the storage associated with the cached psf. The psf can however still be recovered from the Convolver object