#include <otbConfusionMatrixCalculator.h>

Inheritance diagram for otb::ConfusionMatrixCalculator< TRefListLabel, TProdListLabel >:

Collaboration diagram for otb::ConfusionMatrixCalculator< TRefListLabel, TProdListLabel >:

Public Types
typedef RefListLabelType::ValueType::ValueType	ClassLabelType

typedef otb::ConfusionMatrixMeasurements< ConfusionMatrixType, ClassLabelType >	ConfusionMatrixMeasurementsType

typedef itk::VariableSizeMatrix< unsigned long >	ConfusionMatrixType

typedef itk::SmartPointer< const Self >	ConstPointer

typedef std::map< ClassLabelType, int >	MapOfClassesType

typedef std::map< int, ClassLabelType >	MapOfIndicesType

typedef itk::VariableLengthVector< double >	MeasurementType

typedef itk::SmartPointer< Self >	Pointer

typedef ProdListLabelType::Pointer	ProdListLabelPointerType

typedef TProdListLabel	ProdListLabelType

typedef RefListLabelType::Pointer	RefListLabelPointerType

typedef TRefListLabel	RefListLabelType

typedef ConfusionMatrixCalculator	Self

typedef itk::Object	Superclass

Public Member Functions
void	Compute (void)

virtual ::itk::LightObject::Pointer	CreateAnother (void) const

MapOfClassesType	GetMapOfClasses () const

MapOfIndicesType	GetMapOfIndices () const

virtual const char *	GetNameOfClass () const


virtual void	SetReferenceLabels (RefListLabelType *_arg)

virtual const RefListLabelType *	GetReferenceLabels () const

virtual void	SetProducedLabels (ProdListLabelType *_arg)

virtual const ProdListLabelType *	GetProducedLabels () const

virtual MeasurementType	GetTruePositiveValues ()

virtual MeasurementType	GetTrueNegativeValues ()

virtual MeasurementType	GetFalsePositiveValues ()

virtual MeasurementType	GetFalseNegativeValues ()

virtual double	GetTruePositiveValue ()

virtual double	GetTrueNegativeValue ()

virtual double	GetFalsePositiveValue ()

virtual double	GetFalseNegativeValue ()

virtual double	GetKappaIndex ()

virtual double	GetOverallAccuracy ()

virtual MeasurementType	GetPrecisions ()

virtual MeasurementType	GetRecalls ()

virtual MeasurementType	GetFScores ()

virtual double	GetPrecision ()

virtual double	GetRecall ()

virtual double	GetFScore ()

virtual unsigned short	GetNumberOfClasses ()

virtual unsigned long	GetNumberOfSamples ()

virtual ConfusionMatrixType	GetConfusionMatrix ()

Static Public Member Functions
static Pointer	New ()

Protected Member Functions
	ConfusionMatrixCalculator ()

void	PrintSelf (std::ostream &os, itk::Indent indent) const override

	~ConfusionMatrixCalculator () override

Private Member Functions
	ConfusionMatrixCalculator (const Self &)=delete

void	operator= (const Self &)=delete

Private Attributes
ConfusionMatrixMeasurementsType::Pointer	m_ConfMatMeasurements

ConfusionMatrixType	m_ConfusionMatrix

double	m_FalseNegativeValue

MeasurementType	m_FalseNegativeValues

double	m_FalsePositiveValue

MeasurementType	m_FalsePositiveValues

double	m_FScore

MeasurementType	m_FScores

double	m_KappaIndex

MapOfClassesType	m_MapOfClasses

MapOfIndicesType	m_MapOfIndices

unsigned short	m_NumberOfClasses

unsigned long	m_NumberOfSamples

double	m_OverallAccuracy

double	m_Precision

MeasurementType	m_Precisions

ProdListLabelPointerType	m_ProducedLabels

double	m_Recall

MeasurementType	m_Recalls

RefListLabelPointerType	m_ReferenceLabels

double	m_TrueNegativeValue

MeasurementType	m_TrueNegativeValues

double	m_TruePositiveValue

MeasurementType	m_TruePositiveValues

Detailed Description

template<class TRefListLabel, class TProdListLabel>
class otb::ConfusionMatrixCalculator< TRefListLabel, TProdListLabel >

This class computes a confusion matrix from 2 lists of labels. It assumes that the 2 lists have the same length and uses the position of the labels in the lists to build the pairs reference/produced labels.

The rows and columns of the output confusion matrix are sorted according to increasing class labels.

For a 2 classes problem, the confusion matrix is organized as follows:

$\left( \begin{array}{cc} True Positives & False Negatives \\ False Positives & True Negatives \end{array} \right)$

Please note that when accessing the confusion matrix values, the first index is the row index (reference samples), and the second is the column index (produced samples), so that accessing the false positive rate is done by calling GetConfusionMatrix()[1, 0] for the case of a 2 classes problem.

Some measurements are computed by this class: If we consider true positive (TP), true negative (TN), false positive (FP) and false negative (FP) rates, then in the 2 classes case:

$precision = \frac{TP}{\left( TP + FP \right) }$

$recall = \frac{TP}{\left( TP + FN \right) }$

$FScore = \frac{2 * precision * recall}{\left( precision + recall \right) }$

In case of multiclasses problem, these measurements are extended by considering one class versus others.

Moreover overall accuracy and

$\kappa$

index are computed.

Definition at line 58 of file otbConfusionMatrixCalculator.h.