org.openimaj.math.geometry.shape

Class PointDistributionModel

java.lang.Object

org.openimaj.math.geometry.shape.PointDistributionModel

@Reference(author={"Cootes, T. F.","Taylor, C. J."},title="Statistical Models of Appearance for Computer Vision",type=Unpublished,month="October",year="2001",url="http://isbe.man.ac.uk/~bim/Models/app_model.ps.gz") @Reference(type=Inproceedings,author={"C. J. Taylor","D. H. Cooper","J. Graham"},title="Training models of shape from sets of examples",year="1992",booktitle="Proc. BMVC92, Springer-Verlag",pages={"9","","18"})
public class PointDistributionModel
extends Object

A 2d point distribution model learnt from a set of PointLists with corresponding points (the ith point in each PointList is the same landmark). The pdm models the mean shape and the variance from the mean of the top N principal components. The model is generative and can generate new shapes from a scaling vector. To ensure that newly generated shapes are plausible, scaling vectors have PointDistributionModel.Constraints applied to them.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	PointDistributionModel.BoxConstraint A constraint that ensures that each individual element of the scaling vector is within +/- x standard deviations of the model.
static interface	PointDistributionModel.Constraint Interface for modelling constraints applied to the scaling vector of PointDistributionModels so that generated models are plausible.
static class	PointDistributionModel.EllipsoidConstraint Constrain the scaling vector to a hyper-ellipsoid.
static class	PointDistributionModel.NullConstraint A constraint that does nothing.

Field Summary

Fields

Modifier and Type	Field and Description
protected PointDistributionModel.Constraint	constraint
protected int	maxIter
protected PointList	mean
protected int	numComponents
protected PrincipalComponentAnalysis	pc

Constructor Summary

Constructors

Constructor and Description
PointDistributionModel(List<PointList> data) Construct a PointDistributionModel from the given data with a PointDistributionModel.NullConstraint.
PointDistributionModel(PointDistributionModel.Constraint constraint, List<PointList> data) Construct a PointDistributionModel from the given data and PointDistributionModel.Constraint.

Method Summary

Modifier and Type	Method and Description
IndependentPair<Jama.Matrix,double[]>	fitModel(PointList observed) Determine the best parameters of the PDM for the given model.
PointList	generateNewShape(double[] scaling) Generate a plausible new shape from the scaling vector.
PointList	getMean()
double[]	getStandardDeviations(double multiplier) Compute the standard deviations of the shape components, multiplied by the given value.
void	setNumComponents(int n) Set the number of components of the PDM
void	setNumComponents(PrincipalComponentAnalysis.ComponentSelector selector) Set the number of components of the PDM using a PrincipalComponentAnalysis.ComponentSelector .

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

constraint

protected PointDistributionModel.Constraint constraint

pc

protected PrincipalComponentAnalysis pc

mean

protected PointList mean

numComponents

protected int numComponents

maxIter

protected int maxIter

Constructor Detail

PointDistributionModel

public PointDistributionModel(List<PointList> data)

Construct a PointDistributionModel from the given data with a PointDistributionModel.NullConstraint.

Parameters:

data -

PointDistributionModel

public PointDistributionModel(PointDistributionModel.Constraint constraint,
                              List<PointList> data)

Construct a PointDistributionModel from the given data and PointDistributionModel.Constraint.

Parameters:

constraint -

data -

Method Detail

getMean

public PointList getMean()

Returns:

the mean shape

setNumComponents

public void setNumComponents(int n)

Set the number of components of the PDM

Parameters:

n - number of components

setNumComponents

public void setNumComponents(PrincipalComponentAnalysis.ComponentSelector selector)

Set the number of components of the PDM using a PrincipalComponentAnalysis.ComponentSelector .

Parameters:

selector - the PrincipalComponentAnalysis.ComponentSelector to apply.

generateNewShape

public PointList generateNewShape(double[] scaling)

Generate a plausible new shape from the scaling vector. The scaling vector is constrained by the underlying PointDistributionModel.Constraint before being used to generate the model.

Parameters:

scaling - scaling vector.

Returns:

a new shape

getStandardDeviations

public double[] getStandardDeviations(double multiplier)

Compute the standard deviations of the shape components, multiplied by the given value.

Parameters:

multiplier - the multiplier

Returns:

the multiplied standard deviations

fitModel

public IndependentPair<Jama.Matrix,double[]> fitModel(PointList observed)

Determine the best parameters of the PDM for the given model.

Parameters:

observed - the observed model.

Returns:

the parameters that best fit the model.