org.openimaj.image.feature.local.interest

Class AbstractStructureTensorIPD

java.lang.Object

org.openimaj.image.feature.local.interest.AbstractStructureTensorIPD

All Implemented Interfaces:

Cloneable, InterestPointDetector<InterestPointData>, MultiscaleInterestPointDetector<InterestPointData>

Direct Known Subclasses:

HarrisIPD, HessianIPD, LaplaceIPD, QuadratureIPD, RidgeIPD

public abstract class AbstractStructureTensorIPD
extends Object
implements MultiscaleInterestPointDetector<InterestPointData>

Abstract base class for an interest point detector which uses derivatives or the (multiscale) structure tensor.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk), Sina Samangooei (ss@ecs.soton.ac.uk)

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	AbstractStructureTensorIPD.Maxima

Field Summary

Fields

Modifier and Type	Field and Description
protected int	borderSkip
protected float	detectionScale
protected float	detIntScaleFactor
protected float	integrationScale
FImage	lxmxblur
FImage	lxmyblur
FImage	lymyblur
protected List<AbstractStructureTensorIPD.Maxima>	maxima

Constructor Summary

Constructors

Constructor and Description
AbstractStructureTensorIPD(float detIntScaleFactor) Set the scale factor between the integration scale and the detection scale.
AbstractStructureTensorIPD(float detectionScale, float integrationScale) Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale
AbstractStructureTensorIPD(float detectionScale, float integrationScale, boolean blurred) Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale.
AbstractStructureTensorIPD(float detectionScale, float integrationScale, int borderSkip) Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale.
AbstractStructureTensorIPD(float detectionScale, float integrationScale, int borderSkip, boolean blurred) Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale.

Method Summary

Modifier and Type	Method and Description
AbstractStructureTensorIPD	clone()
abstract FImage	createInterestPointMap()
protected void	detectMaxima(FImage image, Rectangle window)
void	findInterestPoints(FImage image) Find the interest points in an image
void	findInterestPoints(FImage image, Rectangle window) Find the interest points in an image
FValuePixel	findMaximum(Rectangle window)
float	getDetectionScale()
float	getDetIntScaleFactor()
float	getIntegrationScale()
List<InterestPointData>	getInterestPoints() Get all the interest points found.
List<InterestPointData>	getInterestPoints(float threshold) Retrieve the interest points found whose normalised score exceeds the threshold
List<InterestPointData>	getInterestPoints(int npoints) Retrieve the interest points found
List<InterestPointData>	getInterestPointsThresh(float thresh)
Jama.Matrix	getSecondMomentsAt(int x, int y)
int	pointsFound()
void	prepareInterestPoints(FImage image)
void	printStructureTensorStats()
void	setDetectionScale(float detectionScale) Set the detection scale for the detector
void	setDetIntScaleFactor(float detIntScaleFactor)
void	setImageBlurred(boolean blurred)
void	setIntegrationScale(float integrationScale)

Methods inherited from class java.lang.Object

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

Field Detail

borderSkip

protected int borderSkip

lxmxblur

public FImage lxmxblur

lymyblur

public FImage lymyblur

lxmyblur

public FImage lxmyblur

detectionScale

protected float detectionScale

integrationScale

protected float integrationScale

detIntScaleFactor

protected float detIntScaleFactor

maxima

protected List<AbstractStructureTensorIPD.Maxima> maxima

Constructor Detail

AbstractStructureTensorIPD

public AbstractStructureTensorIPD(float detIntScaleFactor)

Set the scale factor between the integration scale and the detection scale. When detection scale is set, integration scale = detIntScaleFactor * detectionScale

Parameters:

detIntScaleFactor -

AbstractStructureTensorIPD

public AbstractStructureTensorIPD(float detectionScale,
                                  float integrationScale)

Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale

Parameters:

detectionScale -

integrationScale -

AbstractStructureTensorIPD

public AbstractStructureTensorIPD(float detectionScale,
                                  float integrationScale,
                                  boolean blurred)

Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale. Also state whether the image in from which features are extracted is already blurred to the detection scale, if not it will be blurred to the correct level

Parameters:

detectionScale -

integrationScale -

blurred -

AbstractStructureTensorIPD

public AbstractStructureTensorIPD(float detectionScale,
                                  float integrationScale,
                                  int borderSkip)

Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale. Also specify how many pixels to skip around the edge of the image. The kernel used to extract edges results in a black border so some pixels are better ignored in terms of corner detection.

Parameters:

detectionScale -

integrationScale -

borderSkip -

AbstractStructureTensorIPD

public AbstractStructureTensorIPD(float detectionScale,
                                  float integrationScale,
                                  int borderSkip,
                                  boolean blurred)

Abstract structure tensor detected at a given scale, the first derivatives found and a structure tensor combined from these first derivatives with a gaussian window of sigma = integrationScale. Also specify how many pixels to skip around the edge of the image. The kernel used to extract edges results in a black border so some pixels are better ignored in terms of corner detection. Also state whether the image in from which features are extracted is already blurred to the detection scale, if not it will be blurred to the correct level

Parameters:

detectionScale -

integrationScale -

borderSkip -

blurred -

Method Detail

prepareInterestPoints

public void prepareInterestPoints(FImage image)

printStructureTensorStats

public void printStructureTensorStats()

findInterestPoints

public void findInterestPoints(FImage image)

Description copied from interface: InterestPointDetector

Find the interest points in an image

Specified by:

findInterestPoints in interface InterestPointDetector<InterestPointData>

findInterestPoints

public void findInterestPoints(FImage image,
                               Rectangle window)

Description copied from interface: InterestPointDetector

Find the interest points in an image

Specified by:

findInterestPoints in interface InterestPointDetector<InterestPointData>

findMaximum

public FValuePixel findMaximum(Rectangle window)

detectMaxima

protected void detectMaxima(FImage image,
                            Rectangle window)

createInterestPointMap

public abstract FImage createInterestPointMap()

getInterestPoints

public List<InterestPointData> getInterestPoints(int npoints)

Description copied from interface: InterestPointDetector

Retrieve the interest points found

Specified by:

getInterestPoints in interface InterestPointDetector<InterestPointData>

Parameters:

npoints - number of interest points to retrieve, < 0 returns all

Returns:

interest points

getDetIntScaleFactor

public float getDetIntScaleFactor()

setDetIntScaleFactor

public void setDetIntScaleFactor(float detIntScaleFactor)

getDetectionScale

public float getDetectionScale()

setImageBlurred

public void setImageBlurred(boolean blurred)

setDetectionScale

public void setDetectionScale(float detectionScale)

Description copied from interface: MultiscaleInterestPointDetector

Set the detection scale for the detector

Specified by:

setDetectionScale in interface MultiscaleInterestPointDetector<InterestPointData>

Parameters:

detectionScale - the variance of the Gaussian

getIntegrationScale

public float getIntegrationScale()

setIntegrationScale

public void setIntegrationScale(float integrationScale)

getInterestPoints

public List<InterestPointData> getInterestPoints()

Description copied from interface: InterestPointDetector

Get all the interest points found.

Specified by:

getInterestPoints in interface InterestPointDetector<InterestPointData>

Returns:

all interest points

getInterestPoints

public List<InterestPointData> getInterestPoints(float threshold)

Description copied from interface: InterestPointDetector

Retrieve the interest points found whose normalised score exceeds the threshold

Specified by:

getInterestPoints in interface InterestPointDetector<InterestPointData>

Parameters:

threshold - normalised threshold

Returns:

interest points

getInterestPointsThresh

public List<InterestPointData> getInterestPointsThresh(float thresh)

getSecondMomentsAt

public Jama.Matrix getSecondMomentsAt(int x,
                                      int y)

clone

public AbstractStructureTensorIPD clone()

Overrides:

clone in class Object

pointsFound

public int pointsFound()