Proposals:Refactoring Statistics Framework 2007 Action Items

Action Items

API Fixes

• ImageToCoocurrenceListAdaptor
  • Create a Filter for this operation, in this class the Compute() method will be used.
  • Fix the API so it is a real Adaptor : must have a GetMeasurementVector(unsigned int id) method.
• There is no conceptual difference between "Generators" and "Calculators"
  • They should become Filters
• Estimators have similar characterstics as "Generators" and "Calculators"
  • They could be converted to filters

Proposals

• Sample class could be derived from a DataObject
  • Subsequently, all the derived classes such as ListSampleBase, Histogram and Subsample will be part of the pipeline.
• Add a typedef in the Sample class for the DataObjectDectorator of a Measurement vector
• KDTree could be derived from a DataObject
• ListSampleBase : to be deprecated
• SampleAlgorithmBase will be derived from ProcessObject
  • Subsequently, classes derived from the SampleAlgorithmBase will be process objects.
• The following Calculator classes will also be derived from process object
  • ScalarImageTextureCalculator
  • GreyLevelCooccurrenceMatrixTextureCoefficientsCalculator
• The following Generator classes will be derived from process object
  • ImageToHistogramGenerator
  • ImageToListGenerator
  • KdTreeGenerator
  • ListSampleToHistogramGenerator
  • MaskedScalarImageToGreyLevelCooccurrenceMatrixGenerator
  • MembershipSampleGenerator
  • ScalarImageToGreyLevelCooccurrenceMatrixGenerator
  • ScalarImageToHistogramGenerator
  • SelectiveSubsampleGenerator
  • WeightedCentroidKdTreeGenerator
• The following two generator classes will remain as generators
  • NormalVariateGenerator
  • RandomVariateGeneratorBase
• Frequency container classes ( DenseFrequencyContainer and SparseFrequencyContainer ) do not need any modification at this point unless
  • there is a filter which outputs this data type ( in which case DataObjectDecorator can be defined )
• There is a duplication of data in the omponentbase, membership function

Progress of Specific Tasks

Data Objects

• Sample (Luis) Done
• ListSample (Luis) Done
• Histogram (Luis) Done
• Subsample (Andinet) Done
• MembershipSample (Luis) In Progress
• KdTree (Andinet) Done
  • Replace "WeightedCentroid" with "SumMeasurmentVector"
  • propagate this to KdTree generators
• VariableDimensionHistogram (Luis) Done
• MembershipFunctionBase (Andinet) Done

Process Objects

• MeanFilter (Andinet) Done!
• SampleToHistogramFilter (it was ListSampleToHistogramGenerator) (Luis) In Progress
• ImageToCooccurrenceListFilter (it was ImageToCooccurrenceListAdaptor) (Luis)
• CovarianceFilter (Andinet)
• HistogramToTextureFeaturesFilter (it was GreyLevelCooccurrenceMatrixTextureCoefficientsFilter) (Andinet) Done!
• ScalarImageToTextureFeaturesFilter (Andinet)
• WeightedCovarianceFilter (Andinet)
• WeightedMeanFilter (Andinet) Done!
• ImageToHistogramFilter (was ImageToHistogramGenerator) (Luis)
• ImageToListSampleFilter (ImageToListGenerator) (Andinet) Done!
• SampleToKdTreeFilter (it was KdTreeGenerator ) (Andinet)
  • (+option for generating sum vectors)
  • WeigthedCentroidKdTreeGenerator will be deprecated
• ScalarImageToCooccurrenceMatrixFilter (it was ScalarImageToGreyLevelCooccurrenceMatrixGenerator)(Andinet) Done!
  • MaskedScalarImageToGreyLevelCooccurrenceMatrixGenerator will be deprecated
  • ScalarImageToCooccurrenceMatrixFilter takes a mask image if provided
• ScalarImageToHistogramFilter (it was ScalarImageToHistogramGenerator)(Luis)
• SampleToSubsampleFilter (New class) (Luis) In Progress
  • ClassLabelSubsamplerFilter (it was SelectiveSubsampleGenerator) (Luis)
  • NeighborSubsamplerFilter (it was NeighborhoodSampler)(Luis) In Progress

• ListSampleToHistogramFilter (make it derive from ProcessObject) Andinet
• SampleFilterBase (it was SampleAlgorithmBase) (make it derive from ProcessObject) (Luis)
• HistogramToImageFilter (remove decorator, now that Histogram is a DataObject) (Luis)
  • HistogramToEntropyImageFilter (do code review) (Andinet)
  • HistogramToIntensityImageFilter (do code review) (Luis)
  • HistogramToLogProbabilityImageFilter (do code review) (Andinet)
  • HistogramToProbabilityImageFilter (do code review) (Luis)
• SampleToHistogramProjectionFilter (output Histogram should be created in MakeOutput)(Andinet)
• MeanShiftBlurringSampleFilter (it was SampleMeanShiftBlurringFilter)(Andinet)
• SelectiveMeanShiftBlurringSampleFilter (it was SampleSelectiveMeanShiftBlurringFilter)(Luis)
• MeanShiftClusteringSampleFilter (it was SampleMeanShiftClusteringFilter) (Andinet)
• SelectiveMeanShiftClusteringSampleFilter (it was SampleSelectiveMeanShiftClusteringFilter) (Luis)(why should be called "Selective")
• SampleClassifierFilter (it was SampleClassifier) (Luis) In Progress
• SampleClassifierWithMaskFilter (it was SampleClassifier) (Luis)
• ExpectationMaximizationMixtureModelEstimator (Luis)
  • (review the Classifiers hierarchy)
  • Review Components and MembershipSample classes
• SampleToKmeansFilter (it was KdTreeBasedKmeansEstimator) (Andinet)
  • (it now should use the WeightedKdTreeGenerator inside).
  • Output should be a container of MeasuremenVectors of RealType (Andinet)
  • Rework the Insight/Code/Algorithms/itkScalarImageKmeansImageFilter accordingly (Luis)
• ClassificationSampleFilter (it was SampleClassifier) (Andinet)
  • Related to the classifier reorganization
• MaskedClassificationSampleFilter (it was SampleClassifierWithMask) (Luis)
  • Related to the classifier reorganization

Adaptors

• ImageToListAdaptor (Andinet)
• JointDomainImageToListAdaptor (Andinet)
• PointSetToListAdaptor (Luis)
• ScalarImageToListAdaptor (Andinet)

Classifier related classes refactoring

• Reorganize distance metrics and membership function bases as suggested in [[1]] (Andinet :Done)
  • Use the new MeasurmentVectorTraits added to set/get MeasurementVectorSize
  • Update the migration guide ([[2]])
  • Add unit tests if necessary
  • Fix code style issues
• Refactor Gaussian density function
  • Rename itkGaussianDensityfunction as GaussianMembership function
  • Remove itkDenistyFunction (it does exactly what itkMembershipfunction is supposed to do)