Proposals:Refactoring Statistics Framework 2007 Migration Users Guide

How To Migrate your code to the New Statistics Framework

Measurement Containers



Class Name	Fate	Deprecated API	New API
Sample	Refactored	Does not derives from Object anymore, now it derives from DataObject. typedef for InstanceIdentifier was replaced from unsigned long to std::vector<MeasurementVectorType>::size_type Size() method now returns InstanceIdentifier type, before it returned unsigned int, which was inconsistent with the declaration of InstanceIdentifier as unsigned long. The return type of Size() must be the same use for Id of the elements, since it may reach the same maximum number. Removing unnecessary const in identifiers GetMeasurementVector( const InstanceItentifier & id ) is now GetMeasurementVector( InstanceItentifier id ) GetFrequency( const InstanceItentifier & id ) is now GetFrequency( InstanceItentifier id ) SetMeasurementVectorSize( const MeasurementVectorSizeType s ) is now SetMeasurementVectorSize( MeasurementVectorSizeType s )	const-correctness is now enforced. It now derives from DataObject
Subsample	Refactored	The concept of Indices and InstanceIdentifiers was fixed. InstanceIdentifiers were previously referring to the original Sample, they are now referring to the current Subsample set of measurement vectors. It now derives from its template argument. Before, it was deriving from an explicit itk::Statistics::Sample<> whose template argument MeasurementVector was taken as a trait from the Subsample template argument. The current derivation is more consistent with the notion that the Subsample should behave as the sample from which the instances are taken. FrequencyType GetFrequencyByIndex(int index) const : was removed. This method becomes obsolete when the inconsistencies between index and identifier are removed. InstanceIdentifier GetInstanceIdentifier(int index) const; was removed. This method becomes obsolete when the inconsistencies between index and identifier are removed. Iterators Common GetInstanceIdentifier() now returns the integer distance to the Begin() element. Before it used to return the index of the instance in the real sample. Iterator& operator+(int n) and Iterator& operator-(int n) were removed. The methods were not used, and they were not consistent with other containers. They were also unsafe, given that with Begin() and End() there is no way to verify if an iterator has moved before Begin() or after End() (unless we rely on being aware of the underlying pointer implementation of the Iterators, which would be a violation of the class encapsulation). Constructor ConstIterator(typename InstanceIdentifierHolder::const_iterator iter, const Self* classSample) was made protected, since this constructor must only be called by the Sample owner class. ConstIterator removing Friendship to the Iterator non-const Iterator removing Friendship to the ConstIterator	The basic principle followed now in the design of the Subsample class is that the subsample should behave as a sample. That is, a user should be able to use a Subsample class in any algorithm that is expecting a Sample. AddInstance( InsitanceIdentifier Id ) Now throws an exception if the Id is larger than the Size of the Sample. Now it calls Modified() Iterators Now the non-const Iterator derives from the ConstIterator Methods declared protected and purposely not implemented Default constructor: This was a safety risk since it contained uninitialized variables The following method are in this category to enforce const correctness. All of them could have offered ways of removing constness without warning Iterator(const Self * sample); Iterator(typename InstanceIdentifierHolder::const_iterator iter, const Self* classSample); Iterator(const ConstIterator & it); ConstIterator& operator=(const ConstIterator& it);
ListSampleBase	Deprecated. The class was unecessary, it was providing only a Search() method that should in fact be provided by a process class, not a data container.	If you were deriving a class from the itk::Statistics::ListSampleBase, you should now derive directly from the itk::Statistics::Sample class. If you were using the Search() method you can get that functionality now from the (refactored) NeighborhoodSampler filter class. Note that the NeighborhoodSampler now allows users to setup a specific distance metric.
ListSample	Refactored	It does not derive from the ListSampleBase (since that class was deprecated) The iterators API was updated, const-correctness is now enforced Constructor with instance identifier was moved to the protected section ConstIterator does not refer to the non-const Iterator any more. Method that allowed to convert ConstIterator into non-const Iterators were removed because they were facilitating the violation of const-correctness Various API modifications SearchResultVectorType trait was removed. Search is to be implemented as an external algorithm. Resize( unsigned int ) now is Resize( InstanceIdentifier ) (keeping 64bits in mind) PushBack() before passed the argument by copy, now it passes it as a const reference Size() method used to return unsigned int, now it returns InstanceIdentifier (keeping 64bits in mind) GetMeasurementVector() instance identifier argument is now passed by copy, since it is just a number SetMeasurementVector() now passes InstanceIdentifier by copy, not by const reference SetMeasurement() now passes InstanceIdentifier by copy, not by const reference GetFrequency() now passes InstanceIdentifier by copy, not by const reference GetMeasurementVectorSize() method was removed. The default implementation of the superclass is now functional for this class too. The implementation code of most methods was moved to the .txx file.	It derives now from the Sample class directly Iterators ConstIterator use friendship to the ListSample operator-- removed non-const Iterator is no longer a friend of the ConstIterator. Instead it derives from the ConstIterator The following methods were declared but not implemented, in the protected section with the purpose of preventing const-correctness violations Iterator() constructor Iterator(const Self * sample) constructor of non-const iterator from const sample Iterator( typename InternalDataContainerType::const_iterator iter, InstanceIdentifier iid) constructor of non-const iterator from const sample iterator Iterator(const ConstIterator & it) constructor of non-const iterator from ConstIterator ConstIterator& operator=(const ConstIterator& it) Assignment of ConstIterator to non-const Iterator. The followin methods were moved to the protected section, since only the ListSample class (which is a friend) should call them Iterator( typename InternalDataContainerType::iterator iter, InstanceIdentifier iid):ConstIterator( iter, iid )
MembershipSample	Refactoring in Progress Luis	It does not derive from itk::Sample anymore, instead it derives from DataObject.
Histogram	Refactored same name	Modified by removing the measurement vector size from its template parameters. This Histogram class can now be used for Samples whose measurement vector size is not known at compilation time. The consequence is that now users MUST remember to set the measurement vector size before they call the Initialize() method, otherwise, an Exception will be thrown. GetIndex() now take the InstanceIdentifier by copy, not as a const reference. Size() doesn't return unsigned int anymore. Instead it returns InstanceIdentifier (64 bits in mind) GetSize() now returns a const reference to the Size object. GetSize( dimension ) : const specifier of the argument was removed (it was useless) GetBinMin( ), GetBinMax(), SetBinMin and SetBinMax() const specifier of the dimension was removed (it was useless) bin number parameter type "unsigned long" was replaced with InstanceIdentifier GetBinMinFromValue( ), GetBinMaxFromValue() const specifier of the dimension was removed (it was useless) GetDimensionMins( ), GetDimensionMaxs() const specifier of the dimension was removed (it was useless) GetHistogramMinFromIndex() and GetHistogramMaxFromIndex() Is now declared const (as any Get method should be) It returns a const reference (it used to return a non-const reference) GetFrequency() const specifier of the InstanceIdentifier was removed (it was useless) SetFrequency( value ) const specifier of the FrequencyType value was removed (it was useless) SetFrequency( id, value ) const reference specifier of the InstanceIdentifier was removed. It is now passed by copy const specifier of the FrequencyType value was removed (it was useless) IncreaseFrequency( id, value ) const reference specifier of the InstanceIdentifier was removed. It is now passed by copy const specifier of the FrequencyType value was removed (it was useless) IncreaseFrequency( index, value ) const specifier of the FrequencyType value was removed (it was useless) GetMeasurementVector( id ) const specifier of the InstanceIdentifier was removed (it was useless) GetMeasurement( id, dimension ) type "const unsigned long" of the id, was replaced with InstanceIdentifier, (const part was removed, it was useless) const specifier of the dimension value was removed (it was useless) GetFrequency( id, dimension ) type "const unsigned long" of the id, was replaced with InstanceIdentifier, (const part was removed, it was useless) const specifier of the dimension value was removed (it was useless) Quantile( dimension, value ) const specifier of the dimension type was removed (it was useless) const & specifier of the value type was removed (it was useless), value is now passed by copy. SetMeasurementVectorSize() const modified of the argument was removed (it was useless) Iterators non-const Iterator now derives from the ConstIterator SetFrequency() method was removed from the non-const Iterator, and put in the ConstIterator	Iterators API now made compliant with the other containers. const correctness is enforced non-const Iterator derives from the ConstIterator Preventing any construction of non-const Iterator from ConstIterators by making the following methods be "protected" Iterator(const Self * histogram); Iterator(InstanceIdentifier id, const Self * histogram); Iterator(const ConstIterator & it); ConstIterator& operator=(const ConstIterator& it); Constructor with instance identifier was moved out of the public API into the protected section This method is quite unsafe, and doesn't provide any required functionality (it seems to be just a sign of feature-creep)
VariableDimensionHistogram	Merged with the Histogram class above
KdTree	moved with the same name

Frequency Containers

The use of the data type "FrequencyType" defined in the frequency containers was very confusing in the previous statistics framework. FrequencyType was used to refer to the number of counts(hits) in a histogram bin. This is basically what absolute frequency is. In other instances, FrequencyType was used to refer to relative (normalized) frequencies. This concept mixing creates a lot of confusion. Hence, in the new statistics framework, we have separated the two concepts and defined separate types.

//Absolute Frequency type alias
typedef MeasurementVectorTraits::AbsoluteFrequencyType AbsoluteFrequencyType;

//Relative Frequency type alias
typedef MeasurementVectorTraits::RelativeFrequencyType RelativeFrequencyType;

Formal definition of relative and absolute frequencies can be found at

           http://en.wikipedia.org/wiki/Frequency_(statistics)

The above frequency types are defined in the MeasurmentVectorTraits as follows
/** In the old framework, the FrequencyType is set to float. The problem is for large histograms the total frequency can be more than 1e+7, than increasing the frequency by one does not change the total frequency (because of lack of precision). Using double type will also ultimately fall into the same problem. Hence in the new statistics framework, InstanceIdentifier/FrequencyTypes are set to the the largest possible integer on the machine */

typedef std::vector<int>::size_type InstanceIdentifier;

/** Type defined for representing the frequency of measurement vectors */
typedef InstanceIdentifier AbsoluteFrequencyType;
typedef NumericTraits< AbsoluteFrequencyType >::RealType RelativeFrequencyType;
typedef NumericTraits< AbsoluteFrequencyType >::AccumulateType TotalAbsoluteFrequencyType;
typedef NumericTraits< RelativeFrequencyType >::AccumulateType TotalRelativeFrequencyType;

The containers are renamed with a suffix "2" to be able to generate a separate ITKStatistics library and be able to link with the old library during the refactoring. But once the refactoring is finialized, the name should be changed back to the original names.

Class Name	Fate	Deprecated API	New API
DenseFrequencyContainer	Refactored into DenseFrequencyContainer2	FrequencyType typedef is deprecated ( replaced with AbsoluteFrequencyType and RelativeFrequencyType ) TotalFrequencyType typedef is deprecated ( replaced with AbsoluteFrequencyType and RelativeFrequencyType )	New data types are defined AbsoluteFrequencyType RelativeFrequencyType AbsoluteTotalFrequencyType RelativeTotalFrequencyType
SparseFrequencyContainer	Refactored into SpareFrequencyContainer2	FrequencyType typedef is deprecated ( replaced with AbsoluteFrequencyType and RelativeFrequencyType ) TotalFrequencyType typedef is deprecated ( replaced with AbsoluteFrequencyType and RelativeFrequencyType )	New data types are defined AbsoluteFrequencyType RelativeFrequencyType AbsoluteTotalFrequencyType RelativeTotalFrequencyType

Traits

Class Name	Fate	Deprecated API	New API
MeasurementVectorTraits	Refactored (same name)		Added Assert() instantiations for comparing length of FixedArray raw pointer, std::vector raw pointer FixedArray reference, std::vector reference Added Assign() method for generalizing the assignment of scalars and vector types Assign() vectorType to vectorType Assign() scalarType to FixedArray< scalarType, 1 > Added trait MeasurementVectorType to be used as MeasurementVectorPixelTraits< PixelType >::MeasurementVectorType This trait generalizes the mechanism for creating a measurement vector using as component type the pixel type of an image Added explicit testing up to 100% code coverage

MeanShift

Class Name	Fate	Deprecated API	New API

Adaptors

Class Name	Fate	Deprecated API	New API
ImageToCoocurrenceListAdaptor	Refactored into "ScalarImageToCoocurrenceListSampleFilter"	Compute() method removed ( Update() ) should be used instead SetImage() method removed ( SetInput() ) should be used instead	Derives from Process Object and integrated into the pipeline SetInput() method added User should invoke Update() to generate the coocurrence list sample GetOutput() returns the list sample generated
ImageToListAdaptor	Refactored into "ImageToListSampleAdaptor". This class handles images with scalar, fixed size array (itk::FixedArray) and variable length vector (itk::VariableLengthVector) pixel types.	The class is not derived from ListSampleBase any more since ListSampleBase has been deprecated The second template parameter ( MeasurementVectorType ) is removed since it is redundant. The measurement vector type is determined from the input Image type using the newly added MeasurementVectorPixelTraits. GetUseBuffer() : This method is renamed as GetUsePointContainer(). The term "buffer" is misleading. This flag is used in the class to differentiate between accessing the pixel value directly from the pixel container or using GetPixel(Index) method after computing the index for the given id using ComputeIndex() method. In addition to the Get method, Set methods have been added as indicated in the new API column.	The class derives from ListSample with the measurement vector type determined from the input image type The Iterator and ConstIterator API is modified to be consistent with the iterator and constiterators of the ListSample class. Added SetUsePixelContainer( bool ) : To turn on and off the UsePixelContainer flag. Added UsePixelContainerOn(), UsePixelContainerOff() : convenient methods to turn on/off the UsePixelContainer flag.
ScalarImageToListAdaptor	Deprecated ( This class is redundant as the new ImageToListSampleAdaptor also handles scalar images with scalar pixel type. )
JointDomainImageToListAdaptor	Refactored into "JointDomainImageToListSampleAdaptor"	The class is not derived from the ImageToListSampleAdaptor. This design decision was made due to the new changes in the ImageToListSample adaptor to handle different pixel types ( scalar, Fixed length array, variable length vector). Search() (and helper method ComputeRegion()) methods are removed . This class is not the appropriate place for such functionality.	Derives from ListSample Iterator and ConstIterator API added and it is consistent with the ListSample class. Set/GetUsePixelContainer() and UsePixelContainerOn/Off methods added
PointSetToListAdapor	Refactored into "PointSetToListSampleAdaptor"	The class is not derived from ListSampleBase any more since ListSampleBase has been deprecated	The class derives from ListSample with the measurement vector type determined from the point set type. GetPointSet() returns a const PointSet ( constant correctness GetSize() returns InstanceIdentifer type ( to have a consistent signature with the superclass ) The Iterator and ConstIterator API is modified to be consistent with the iterator and constiterators of the ListSample class.

Univariate Distributions

Class Name	Fate	Deprecated API	New API

Algorithms

Algorithms, including FloorLog, MedianOfThree, FindSampleBound, FindSampleBoundAndMean, Partition, QuickSelect, NthElement, InsertSort, DownHeap, HeapSort, IntrospectiveSortLoop, IntrospectiveSort, have been moved from the itk::Statistics namespace to the itk::Statistics::Algorithm namespace.

Class Name	Fate	Deprecated API	New API

Calculators

Class Name	Fate	Deprecated API	New API
MeanCalculator	Refactored into MeanSampleFilter	SetSample() method replaced with SetInput() GetOutput() method return type modified ( the new signature returns a data object decorator )	Derives from ProcessObject and the following methods added to integrate the class into the pipeline SetInput() GetOutput() (returns a data object decorated type of the mean vector ) GetMean() (returns the mean vector as it is without any decoration )
WeightedMeanCalculator	Refactored into WeightedMeanSampleFilter	SetSample() method replaced with SetInput() SetWeights() method input argument modified ( the new signature takes dataobject decorated weight array) SetWeightFunction() method replaced with SetWeightingFunction() GetWeights() method return type modified ( new signature returns a data object decorator of the weight array) GetWeightingFunction() method replaced with GetWeightingFunction() GetOutput() method return type modified ( the new signature returns a data object decorator of the mean )	Derives from ProcessObject and the following methods added to integrate the class into the pipeline SetInput() GetOutput() (returns a data object decorated type of the mean vector ) SetWeights() (Takes dataObjecte decorated version of weight array type ) SetWeightingFunction (Takes dataObject decorated version of weighting function type )
CovarianceCalculator	Refactored into CovarianceSampleFilter	SetMean() removed ( the mean will always be computed inside the class. Keeping track of mean computed internally and set from outside creates lots of inconsistencies in the pipeline). SetSample() replaced with SetInput() GetOutput() replaced with GetCovarianceMatrixOutput() which returns a decorated version of the covariance matrix	Derived from process object and the following methods added to integrate the class into the pipeline SetInput() GetMeanOutput() ( returns a decorated version of the mean vector ) GetCovarianceMatrixOutput() ( returns a decorated version of the covariance matrix )
WeightedCovarianceCalculator	Refactored into WeightedCovarianceSampleFilter	SetSample() method replaced with SetInput() SetWeights() method input argument modified ( the new signature takes dataobject decorated weight array) SetWeightFunction() method replaced with SetWeightingFunction() GetWeights() method return type modified ( new signature returns a data object decorator of the weight array) GetWeightingFunction() method replaced with GetWeightingFunction()	Derives from itk::CovarianceFilter and all the methods defined in the CovarianceFilter for pipeline integration (SetInput(), GetMeanOutput(), GetCovarianceMatrixOutput() )should be used SetWeights() (Takes dataObjecte decorated version of weight array type ) SetWeightingFunction (Takes dataObject decorated version of weighting function type ) GetWeights() returns decorated version of the weight array GetWeightingFunction() returns decorated version of the weighting function
GreyLevelCooccurrenceMatrixTextureCoefficientsCalculator	Refactored into HistogramToTextureFeaturesFilter	Compute() removed ( Update() method should be used in the new API ) SetHistogram() removed ( replaced by SetInput() in the new API )	Derives from ProcessObject and the following methods added to integrate it to the pipeline SetInput() GenerateData() ( user should invoke Update() to compute the features ) Data object decorated versions of the Get methods for each feature was added. GetEnergyOutput() GetEntropyOutput() GetCorrelationOutput() GetInverseDifferenceMomentOutput() GetInertiaOutput() GetClusterShadeOutput() GetClusterProminenceOutput() GetHaralickCorrelationOutput()
ScalarImageTextureCalculator	Refactored into ScalarImageToTextureFeaturesFilter	Compute() removed ( Update() method should be used in the new API )	Derives from ProcessObject and the following methods added to integrate it to the pipeline GenerateData() ( user should invoke Update() to compute the features ) Decorated data types for the mean and standard deviation feature values were defined. Data object decorated versions of the GetOutput methods for the mean and standard deviation were added. GetFeatureMeansOutput() GetStandardDeviationsOutput() GetInput and GetMaskImage() methods were added

Generators

Class Name	Fate	Deprecated API	New API
ListSampleToHistogramGenerator	Refactored into SampleToHistogramFilter	It used to derive from Object, now derives from ProcessObject It used to be templated over SampleType and the HistogramMeasurementType Histogram frequency container. Not it is templated over TSample and THistogram SetListSample() removed. Replaced with SetInput() The following member variables were removed and now are managed through Decorated inputs Histogram Size Marginal scale Histogram Mins Histogram Maxs AutoMinMax	It derives now from ProcessObject It is now templated over TSample and THistogram types SetInput( const SampleType * ) const SampleType * GetInput() const Copy Constructor and Operator=() are declared protected are purposely not implemented in order to enforce the correct use of SmartPointers. MakeOutput() method added, so that the filter is compatible with the DisconnectPipeline() API of the output DataObject. Added DataDecorated Inputs for Histogram Size AutoMinimumMaximum MarginalScale HistogramBinMinimum HistogramBinMaximum Decorated inputs provide Set/Get methods for the decorated type and ONLY Set method for the undecorated type. Get method for undecorated type cannot be provided because their API uses a const reference, and the decorated input may have been a NULL pointer, in which case there is not possible const reference to return.
ImageToListGenerator	Refactored into ImageToListSampleFilter	GetListSample() replaced with GetOutput() GenerateData() moved to protected interface	Integrated with the pipeline GetOutput() added The measurement vector type is determined using MeasurementVectorTraits and hence this adaptor handles scalar, fixed array and variable length vector pixel type
ScalarImageToGreyLevelCooccurrenceMatrixGenerator	Refactored into ScalarImageToCooccurrenceMatrixFilter	Compute() method removed ( Update() ) should be used instead	Derives from Process Object and integrated into the pipeline User should invoke Update() to compute the matrix GetOutput() returns the computed matrix Takes a mask image as well SetMaskImage() added GetMaskImage() added
MaskedScalarImageToGreyLevelCooccurrenceMatrixGenerator	Deprecated ( Use ScalarImageToCooccurrenceMatrixFilter instead since this takes a mask image in addition to the input image )
ImageToHistogramGenerator	Refactored into ImageToHistogramFilter	It used to derive from Object, now derives from ProcessObject It used to expect a ListSample as input, it now takes any class that complies with the Sample API Compute() implementation removed, the default Update() of the superclass ProcessObject() provides the correct functionality. Compute() was replaced with GenerateData(). Member variables removed	It derives now from ProcessObject SetInput( const SampleType * ) const SampleType * GetInput() const const HistogramType * GetOutput() const Copy Constructor and Operator=() are declared protected are purposely not implemented in order to enforce the correct use of SmartPointers. MakeOutput() method added, so that the filter is compatible with the DisconnectPipeline() API of the output DataObject. Added DataDecorated Inputs for Histogram Size AutoMinimumMaximum MarginalScale HistogramBinMinimum HistogramBinMaximum Decorated inputs provide Set/Get methods for the decorated type and ONLY Set method for the undecorated type. Get method for undecorated type cannot be provided because their API uses a const reference, and the decorated input may have been a NULL pointer, in which case there is not possible const reference to return.

Filters

Class Name	Fate	Deprecated API	New API
ListSampleToHistogramFilter	Refactored as SampleToHistogramFilter	It used to derive from Object, now derives from ProcessObject It used to expect a ListSample as input, it now takes any class that complies with the Sample API SetListSample() removed. The method SetInput() should be used instead. SetHistogram() removed. The resulting histogram is now obtained from the GetOutput() method, according to a standard Filter API. Update() implementation removed, the default Update() of the superclass ProcessObject() provides the correct functionality. Run() method removed. The method Update() in the superclass provides the equivalent functionality. Member variables removed A pointer to the ListSample. This is now managed by the Input of the ProcessObject. A pointer to the output histogram. This is now manged by the Output of the ProcessObject. In the old API it was expected that the user will Construct an instance of the Histogram class Setup in that histogram all the parameters such as: Histogram Size Histogram bin Minimum Histogram bin Maximum AutoMinMax In the new API this setup is done directly in the filter, since the output object is supposed to be created by the filter, and provided as a const object, therefore the users do not have access to it for applying any modifications.	It derives now from ProcessObject SetInput( const SampleType * ) const SampleType * GetInput() const const HistogramType * GetOutput() const Copy Constructor and Operator=() are declared protected are purposely not implemented in order to enforce the correct use of SmartPointers. MakeOutput() method added, so that the filter is compatible with the DisconnectPipeline() API of the output DataObject. Added DataDecorated Inputs for Histogram Size AutoMinimumMaximum MarginalScale HistogramBinMinimum HistogramBinMaximum Decorated inputs provide Set/Get methods for the decorated type and ONLY Set method for the undecorated type. Get method for undecorated type cannot be provided because their API uses a const reference, and the decorated input may have been a NULL pointer, in which case there is not possible const reference to return.
HistogramToImageFilter	HistogramToImageFilter		ImageDimension was added as second template parameter. This is needed because now the number of components in the Histogram class are defined at run time.
HistogramToIntensityImageFilter	HistogramToIntensityImageFilter		ImageDimension was added as second template parameter. See HistogramToImageFilter
HistogramToLogProbabilityImageFilter	HistogramToLogProbabilityImageFilter		ImageDimension was added as second template parameter. See HistogramToImageFilter
HistogramToProbabilityImageFilter	HistogramToProbabilityImageFilter		ImageDimension was added as second template parameter. See HistogramToImageFilter
HistogramToEntropyImageFilter	HistogramToEntropyImageFilter		ImageDimension was added as second template parameter. See HistogramToImageFilter

Multivariate Density Functions

Class Name	Fate	Deprecated API	New API
DensityFunction	Removed ( it is a duplication of what MembershipFunction) does
GaussianDensityFunction	Refactored as GaussianMembershipFunction

Distance Metrics

Class Name	Fate	Deprecated API	New API
DistanceMetric	Refactored (Same name )	Evaluate method with two arguments removed	Derives now from FunctionBase ( not MembershipFunctionBase ) SetMeasurmentVectorSize reimplemented
EuclideanDistance	Refactored as EuclideanDistanceMetric	Evaluate method with two arguments removed (unnecessary for the classification pipeline ) IsWithinRange method removed (unnecessary for the classification pipeline )
MahalanobisDistanceMetric	Refactored same name
EuclideanSquareDistanceMetric	New class		Same API as EuclideanDistance
ManhattanDistanceMetric	New class		Same API as EuclideanDistance

MembershipFunctions

Class Name	Fate	Deprecated API	New API
DistanceToCentroidMembershipFunction	Refactored (Same name )	Set/Get Number of samples method removed ( not used at all )	SetMeasurmentVectorSize reimplemented
GaussianDensityFunction	Refactored as GaussianMembershipFunction

Process Objects

Class Name	Fate	Deprecated API	New API
SampleClassifier	These are now Classifiers, see below
SampleClassifierWithMask	These are now Classifiers, see below

Classifiers

• SetSample becomes SetInput

Class Name	Fate	Deprecated API	New API
ClassifierBase	Deprecated
SampleClassifier	Refactored (as SampleClassifierFilter)		Derives from ProcessObject to integrate the classifier to ITK pipeline Add Datadecorated inputs Classlabels MembershipFunctions Throw an exception if the specified number of class doesn't match with the number of class labels or the number of membership functions specified. Add Set/Get method for DecisionRule
ImageClassifierBase	Refactored (as ImageClassifierFilter)		Derives from ImageToImageFilter to integrate the classifier to ITK pipeline Add Datadecorated inputs Classlabels MembershipFunctions MembershipFunctionsWeightArray DecisionRule Throw an exception if the specified number of class doesn't match with the number of class labels or the number of membership functions specified. Add Set/Get method for DecisionRule GetOutput returns MembershipSampleType

Decision rules

Class Name	Fate	Deprecated API	New API
DecisionRuleBase	Refactored as DecisionRule	* Remove Evaluate(ArrayType ) method	Define MembershipVector type which is of std::vector< double > type and store the membership score Define only ONE Evaluate method that takes MembershipVector type
MaximumDecisionRule	Refactored as MaximumDecisionRule2
MinimumDecisionRule	Refactored as MinimumDecisionRule2
MaximumDecisionRatioRule	Refactored as MaximumDecisionRationRule2

Components

Class Name	Fate	Deprecated API	New API
itkMixtureModelComponentBase	Refactored(same name)	WeightsArrayType * GetWeights() non-const double GetWeight( int ) non-const void SetWeight( int ) void CreateWeightArray() void DeleteWeightArray()	const WeightsArrayType & GetWeights() const double GetWeight( unsigned int ) const void SetWeight( unsigned int ) Removed (not needed now that Weights is not a pointer) Removed (not needed now that Weights is not a pointer)
itkGaussianMixtureModelComponent	Refactored(same name)

Estimators

Class Name	Fate	Deprecated API	New API
itkKdTreeBasedKmeansEstimator	Refactored (same name)
itkExpectationMaximizationMixtureModelEstimator	Refactored (same name)