org.openimaj.image.feature.local.descriptor.gradient

Class SIFTFeatureProvider

java.lang.Object

org.openimaj.image.feature.local.descriptor.gradient.SIFTFeatureProvider

All Implemented Interfaces:

FeatureVectorProvider<OrientedFeatureVector>, GradientFeatureProvider, GradientFeatureProviderFactory, LocalFeatureProvider<OrientedFeatureVector>

@Reference(type=Article,author="David Lowe",title="Distinctive image features from scale-invariant keypoints",year="2004",journal="IJCV",pages={"91","110"},month="January",number="2",volume="60") @Reference(type=Inproceedings,author="David Lowe",title="Object recognition from local scale-invariant features",year="1999",booktitle="Proc. of the International Conference on Computer Vision {ICCV}",pages={"1150","1157"})
public class SIFTFeatureProvider
extends Object
implements GradientFeatureProvider, GradientFeatureProviderFactory

An extractor for SIFT features. SIFT features are basically multiple edge orientation histograms constructed over a spatial grid. Samples added to the SIFT histogram are weighted with a Gaussian based on the distance from the centre of the sampling window. Samples are also blurred across histogram bins in both the spatial and orientation directions. Based on Section 6 of Lowe's IJCV paper

Author:

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

Field Summary

Fields

Modifier and Type	Field and Description
protected float	gaussianSigma
protected int	numOriBins Number of orientation bins in the histograms
protected int	numSpatialBins Number of spatial bins for the x and y directions in the histograms
protected float	patchOrientation
protected float	sigmaSq2 2 times the weighting Gaussian squared (normalised to the patch size in terms of spatial bins)
protected float	valueThreshold Threshold for the maximum allowed value in the histogram
protected float[]	vec

Constructor Summary

Constructors

Constructor and Description
SIFTFeatureProvider() Construct a SIFTFeatureProvider with the default parameters.
SIFTFeatureProvider(int numOriBins, int numSpatialBins) Construct a SIFTFeatureProvider with the provided options.
SIFTFeatureProvider(int numOriBins, int numSpatialBins, float valueThreshold, float gaussianSigma) Construct a SIFTFeatureProvider with the provided options.

Method Summary

Modifier and Type	Method and Description
void	addSample(float x, float y, float gradmag, float gradori) Add a sample to the feature.
OrientedFeatureVector	getFeatureVector() Get the FeatureVector associated with this object.
float	getOversamplingAmount() Get the amount of required over-sampling outside of the unit square patch.
protected void	interpolateSample(float x, float y, float magnitude, float orientation) Spread the sample around the closest bins in the histogram.
GradientFeatureProvider	newProvider() Construct a new GradientFeatureProvider implementation with all its internal settings setup ready for use.
void	setPatchOrientation(float patchOrientation) Set the primary orientation of the sample patch in the image being processed.

Methods inherited from class java.lang.Object

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

Field Detail

numOriBins

protected int numOriBins

Number of orientation bins in the histograms

numSpatialBins

protected int numSpatialBins

Number of spatial bins for the x and y directions in the histograms

valueThreshold

protected float valueThreshold

Threshold for the maximum allowed value in the histogram

sigmaSq2

protected float sigmaSq2

2 times the weighting Gaussian squared (normalised to the patch size in terms of spatial bins)

gaussianSigma

protected float gaussianSigma

vec

protected float[] vec

patchOrientation

protected float patchOrientation

Constructor Detail

SIFTFeatureProvider

public SIFTFeatureProvider()

Construct a SIFTFeatureProvider with the default parameters.

SIFTFeatureProvider

public SIFTFeatureProvider(int numOriBins,
                           int numSpatialBins)

Construct a SIFTFeatureProvider with the provided options.

Parameters:

numOriBins - the number of orientation bins (default 8)

numSpatialBins - the number of spatial bins in each direction (default 4)

SIFTFeatureProvider

public SIFTFeatureProvider(int numOriBins,
                           int numSpatialBins,
                           float valueThreshold,
                           float gaussianSigma)

Construct a SIFTFeatureProvider with the provided options.

Parameters:

numOriBins - the number of orientation bins (default 8)

numSpatialBins - the number of spatial bins in each direction (default 4)

valueThreshold - threshold for the maximum value allowed in the histogram (default 0.2)

gaussianSigma - the width of the Gaussian used for weighting samples, relative to the half-width of the sampling window (default 1.0).

Method Detail

addSample

public void addSample(float x,
                      float y,
                      float gradmag,
                      float gradori)

Description copied from interface: GradientFeatureProvider

Add a sample to the feature. The x and y coordinates are given in terms of a unit index square (i.e. 0<=x<=1 and 0<=y<=1).

Specified by:

addSample in interface GradientFeatureProvider

Parameters:

x - x-coordinate within the unit indexing square

y - y-coordinate within the unit indexing square

gradmag - the gradient magnitude at the given coordinate

gradori - the gradient direction at the given coordinate

interpolateSample

protected void interpolateSample(float x,
                                 float y,
                                 float magnitude,
                                 float orientation)

Spread the sample around the closest bins in the histogram. If there are 4 spatial bins in each direction, then a sample at 0.25 would get added equally to bins [0] and [1] in the x-direction.

Parameters:

x - the normalised x-coordinate

y - the normalised y-coordinate

magnitude - the magnitude of the sample

orientation - the angle of the sample

getFeatureVector

public OrientedFeatureVector getFeatureVector()

Description copied from interface: FeatureVectorProvider

Get the FeatureVector associated with this object.

Specified by:

getFeatureVector in interface FeatureVectorProvider<OrientedFeatureVector>

Specified by:

getFeatureVector in interface GradientFeatureProvider

Returns:

the feature vector.

setPatchOrientation

public void setPatchOrientation(float patchOrientation)

Description copied from interface: GradientFeatureProvider

Set the primary orientation of the sample patch in the image being processed. This might be used by concrete implementations to provide rotation invariance. This method should only be called once by calling code, and the call should be before any calls to GradientFeatureProvider.addSample(float, float, float, float) are made.

Specified by:

setPatchOrientation in interface GradientFeatureProvider

Parameters:

patchOrientation - the actual orientation of the square in the image

newProvider

public GradientFeatureProvider newProvider()

Description copied from interface: GradientFeatureProviderFactory

Construct a new GradientFeatureProvider implementation with all its internal settings setup ready for use.

Specified by:

newProvider in interface GradientFeatureProviderFactory

Returns:

a new GradientFeatureProvider implementation.

getOversamplingAmount

public float getOversamplingAmount()

Description copied from interface: LocalFeatureProvider

Get the amount of required over-sampling outside of the unit square patch. An oversampling of 0.5 would result in methods such as addSample(float, float, float, float) being called with pixels in the range -0.5 to 1.5.

Specified by:

getOversamplingAmount in interface LocalFeatureProvider<OrientedFeatureVector>

Returns:

the amount over sampling above the unit square