#include <ImFusion/CT/Solver/LinearOperator.h>

Abstract linear operator interface with adjoint. More...

Inheritance diagram for LinearOperator:

Detailed Description

Abstract linear operator interface with adjoint.

Represents a linear operator A that can be written as a matrix with both forward and adjoint operations. Forms the basis for projection and backprojection operators in tomographic reconstruction. Supports large-scale implementations with on-the-fly execution to avoid storing matrices in memory. May run modified versions of input expressions to set interpolation modes and other parameters.

Note: There are practically no use-cases where a user should derive directly from this class. This is because this class provides both an ImageMath and a SharedImageSet input version of its main functions, and each of these can be obtained from the other. The two template specializations of the LinearOperatorEvalInputExpr template provide for each of the cases where only one implementation is provided by a user.

Public Types
using	ExprBase = ImageMath::ExprBase

Public Types inherited from Block
using	Input = DataList

using	Output = DataList

Public Member Functions
	LinearOperator (const SharedImageSet domainRef, const SharedImageSet rangeRef=nullptr)
	Constructor with optional references for the domain and codomain (i.e. range) of the operator.

	~LinearOperator () override=0
	Destructor.

virtual Status	apply (const ExprBase &exprIn, const SharedImageSet &in, SharedImageSet &out, const ExprBase *exprOut=nullptr) const =0
	Applies a linear operator \(A\), optionally pre/post-composing with an ImageMath expression.

virtual Status	apply (const SharedImageSet &in, SharedImageSet &out, const ExprBase *exprOut=nullptr) const =0

template<typename ImplType, typename ImplTypeOut>
Status	apply (const ImageMath::ExprBaseTpl< ImplType > &exprIn, const SharedImageSet &in, SharedImageSet &out, const ImageMath::ExprBaseTpl< ImplTypeOut > *exprOut=nullptr) const

template<typename ImplType>
Status	apply (const ImageMath::ExprBaseTpl< ImplType > &exprIn, const SharedImageSet &in, SharedImageSet &out, const ExprBase *exprOut=nullptr) const

virtual Status	applyAdjoint (const ExprBase &exprIn, const SharedImageSet &in, SharedImageSet &out, const ExprBase *exprOut=nullptr) const =0
	Applies the adjoint of a linear operator \(A\), optionally pre/post-composing with an ImageMath expression.

virtual Status	applyAdjoint (const SharedImageSet &in, SharedImageSet &out, const ExprBase *exprOut=nullptr) const =0

template<typename ImplType, typename ImplTypeOut>
Status	applyAdjoint (const ImageMath::ExprBaseTpl< ImplType > &exprIn, const SharedImageSet &in, SharedImageSet &out, const ImageMath::ExprBaseTpl< ImplTypeOut > *exprOut=nullptr) const

template<typename ImplType>
Status	applyAdjoint (const ImageMath::ExprBaseTpl< ImplType > &exprIn, const SharedImageSet &in, SharedImageSet &out, const ExprBase *exprOut=nullptr) const

Status	runBlock (Input &input, Output &output) override
	Applies the operator to the first element of input.

virtual std::unique_ptr< LinearOperator >	clone () const =0
	Factory method.

std::unique_ptr< LinearOperatorAdjointWrapper >	adjoint () const
	Creates an adjoint operator of this.

const SharedImageSet *	domainRef () const
	Getter for m_domainRef, can be nullptr if this is implicitly defined by the operator.

const SharedImageSet *	rangeRef () const
	Getter for m_rangeRef, can be nullptr if this is implicitly defined by the operator.

virtual std::unique_ptr< SharedImageSet >	createDomainSis () const
	Creates a SharedImageSet which is in the range of the operator.

virtual std::unique_ptr< SharedImageSet >	createRangeSis () const
	Creates a SharedImageSet which is in the range of the operator.

Eigen::MatrixXf	extractApplyMatrix () const
	Extracts a dense matrix representing this operator.

Eigen::MatrixXf	extractApplyAdjointMatrix () const
	Extracts a dense matrix representing this operator.

Public Member Functions inherited from Configurable
virtual void	configure (const Properties *p)
	Configure this object instance by de-serializing the given Properties.

virtual void	configuration (Properties *p) const
	Serialize the current object configuration into the given Properties object.

virtual void	configureDefaults ()
	Retrieve the properties of this object, replaces values with their defaults and sets it again.

void	registerParameter (ParameterBase *param)
	Register the given Parameter or SubProperty, so that it will be configured during configure()/configuration().

void	unregisterParameter (const ParameterBase *param)
	Remove the given Parameter or SubProperty from the list of registered parameters.

	Configurable (const Configurable &rhs)

	Configurable (Configurable &&rhs) noexcept

Configurable &	operator= (const Configurable &)

Configurable &	operator= (Configurable &&) noexcept

Protected Attributes
const SharedImageSet *	m_domainRef = nullptr

const SharedImageSet *	m_rangeRef = nullptr

Protected Attributes inherited from Configurable
std::vector< Param >	m_params
	List of all registered Parameter and SubProperty instances.

Additional Inherited Members
Public Attributes inherited from Configurable
Signal	signalParametersChanged
	Emitted whenever one of the registered Parameters' or SubPropertys' `signalValueChanged` signal was emitted.

Member Function Documentation

◆ apply() [1/4]

virtual Status apply	(	const ExprBase &	exprIn,
		const SharedImageSet &	in,
		SharedImageSet &	out,
		const ExprBase *	exprOut = nullptr ) const

pure virtual

Applies a linear operator \(A\), optionally pre/post-composing with an ImageMath expression.

Computes \(exprOut(A * exprIn(in))\).

As \(A\) is linear, it can in principle be represented as a matrix. However we aim at large-scale implementations meaning that the corresponding matrix is typically to large to be kept in memory. We therefore only require an implementation of the matrix-vector product as opposed to an explicit representation.

Parameters

[in]	exprIn	Expression to which the linear operator is applied
[in]	in	Reference input to define the domain of the operator (this is neccessary as exprIn could be scalar).
[in,out]	out	Is overwritten with result of \(exprOut(A * exprIn(in))\).
[in]	exprOut	If set, this expression will be applied to the results after the operator was applied. If not set, treated as identity. `out` must contain ImageMath::Variable which will be substituted with the result of \(A * exprIn(in)\). In principle, this expression may contain `out` even if the values of `out` are then overwritten by this function. However, an operator may split up \(A\) into blocks, in which case the implemented LinearOperator must use checkOutExprAndCreateUpdateExprAndTmpOut to accumulate the result correctly.

Note: This method must be selection aware.

Implemented in ConvolutionFilter, DiagonalOperatorTpl< OperandT >, GlCBCTProjector, IdentityOperator, LinearOperatorAdjointWrapper, LinearOperatorEvalInputExpr< false >, LinearOperatorEvalInputExpr< true >, and LinearOperatorTreeNode.

◆ apply() [2/4]

virtual Status apply	(	const SharedImageSet &	in,
		SharedImageSet &	out,
		const ExprBase *	exprOut = nullptr ) const

pure virtual

See also: LinearOperator::apply(const ExprBase&, SharedImageSet&, SharedImageSet&, const ExprBase*) with the input expression set to the identity applied to in.

Implemented in ConvolutionFilter, FiniteDifferences, LaplaceOperator, LinearOperatorAdjointWrapper, LinearOperatorEvalInputExpr< false >, LinearOperatorEvalInputExpr< true >, and MatrixLinearOperator.

◆ apply() [3/4]

template<typename ImplType, typename ImplTypeOut>

Status apply	(	const ImageMath::ExprBaseTpl< ImplType > &	exprIn,
		const SharedImageSet &	in,
		SharedImageSet &	out,
		const ImageMath::ExprBaseTpl< ImplTypeOut > *	exprOut = nullptr ) const

inline

See also: LinearOperator::apply(const ExprBase&, SharedImageSet&, SharedImageSet&, const ExprBase*) with the input expression

◆ apply() [4/4]

template<typename ImplType>

Status apply	(	const ImageMath::ExprBaseTpl< ImplType > &	exprIn,
		const SharedImageSet &	in,
		SharedImageSet &	out,
		const ExprBase *	exprOut = nullptr ) const

inline

See also: LinearOperator::apply(const ExprBase&, SharedImageSet&, SharedImageSet&, const ExprBase*) with the input expression

◆ applyAdjoint() [1/3]

virtual Status applyAdjoint	(	const ExprBase &	exprIn,
		const SharedImageSet &	in,
		SharedImageSet &	out,
		const ExprBase *	exprOut = nullptr ) const

pure virtual

Applies the adjoint of a linear operator \(A\), optionally pre/post-composing with an ImageMath expression.

Computes \(exprOut(A^\top * exprIn(in))\).

Parameters

[in]	exprIn	Expression to which the adjoint operator is applied.
[in]	in	Reference input defining the domain of the operator (this is neccessary, as exprIn can be a scalar expression).
[in,out]	out	Is overwritten with result of \(exprOut(A * exprIn(in))\).
[in]	exprOut	If set, this expression will be applied to the results after the operator was applied. If not set, treated as identity. `out` is expected to contain ImageMath::Variable which will be substituted with the result of \(A * exprIn(in)\). In principle, this expression may contain `out` even though `out` is overwritten by this function. However, an operator may split up \(A\) into blocks, in which case the implemented LinearOperator must use checkOutExprAndCreateUpdateExprAndTmpOut to accumulate the result correctly.

Note: This method must be selection aware.

Implemented in ConvolutionFilter, DiagonalOperatorTpl< OperandT >, GlCBCTProjector, IdentityOperator, LinearOperatorAdjointWrapper, LinearOperatorEvalInputExpr< false >, LinearOperatorEvalInputExpr< true >, and LinearOperatorTreeNode.

◆ applyAdjoint() [2/3]

virtual Status applyAdjoint	(	const SharedImageSet &	in,
		SharedImageSet &	out,
		const ExprBase *	exprOut = nullptr ) const

pure virtual

See also: LinearOperator::applyAdjoint(const ExprBase&, SharedImageSet&, SharedImageSet&, const ExprBase*) with the input expression set to the identity applied to in.

Implemented in FiniteDifferences, LaplaceOperator, LinearOperatorAdjointWrapper, LinearOperatorEvalInputExpr< false >, LinearOperatorEvalInputExpr< true >, and MatrixLinearOperator.

◆ applyAdjoint() [3/3]

template<typename ImplType, typename ImplTypeOut>

Status applyAdjoint	(	const ImageMath::ExprBaseTpl< ImplType > &	exprIn,
		const SharedImageSet &	in,
		SharedImageSet &	out,
		const ImageMath::ExprBaseTpl< ImplTypeOut > *	exprOut = nullptr ) const

inline

See also: LinearOperator::apply(const ExprBase&, SharedImageSet&, SharedImageSet&, const ExprBase*) with the input expression

◆ runBlock()

Status runBlock	(	Input &	input,
		Output &	output )

overridevirtual

Applies the operator to the first element of input.

Reimplemented from Block.

◆ clone()

virtual std::unique_ptr< LinearOperator > clone ( ) const

pure virtual

Factory method.

Implemented in ConvolutionFilter, DiagonalOperatorTpl< OperandT >, FiniteDifferences, GlCBCTProjector, IdentityOperator, LaplaceOperator, LinearOperatorAdjointWrapper, LinearOperatorTreeNode, and MatrixLinearOperator.

◆ createDomainSis()

virtual std::unique_ptr< SharedImageSet > createDomainSis ( ) const

virtual

Creates a SharedImageSet which is in the range of the operator.

Reimplemented in LinearOperatorTreeNode.

◆ createRangeSis()

virtual std::unique_ptr< SharedImageSet > createRangeSis ( ) const

virtual

Creates a SharedImageSet which is in the range of the operator.

Reimplemented in FiniteDifferences, and LinearOperatorTreeNode.

◆ extractApplyMatrix()

Eigen::MatrixXf extractApplyMatrix ( ) const

Extracts a dense matrix representing this operator.

Note: The method executes the operator on each single pixel in the domain. It is only advised for small operators. Also be aware of the space requirement.

◆ extractApplyAdjointMatrix()

Eigen::MatrixXf extractApplyAdjointMatrix ( ) const

Extracts a dense matrix representing this operator.

Note: The method executes the adjoint of this operator on each single pixel in the domain. It is only advised for small operators. Also be aware of the space requirement.

The documentation for this class was generated from the following file:

ImFusion/CT/Solver/LinearOperator.h

Detailed Description

Public Types

Public Member Functions

Protected Attributes

Additional Inherited Members

Member Function Documentation

◆ apply() [1/4]

◆ apply() [2/4]

◆ apply() [3/4]

◆ apply() [4/4]

◆ applyAdjoint() [1/3]

◆ applyAdjoint() [2/3]

◆ applyAdjoint() [3/3]

◆ runBlock()

◆ clone()

◆ createDomainSis()

◆ createRangeSis()

◆ extractApplyMatrix()

◆ extractApplyAdjointMatrix()