org.openimaj.image.feature.local.detector.dog.extractor

Class OrientationHistogramExtractor

java.lang.Object

org.openimaj.image.feature.local.detector.dog.extractor.OrientationHistogramExtractor

All Implemented Interfaces:

FeatureVectorExtractor<FloatFV,GradientScaleSpaceImageExtractorProperties<FImage>>

public class OrientationHistogramExtractor
extends Object
implements FeatureVectorExtractor<FloatFV,GradientScaleSpaceImageExtractorProperties<FImage>>

Extract an orientation histogram about an interest point. The window size is proportional to the scale of the interest point, and the samples are weighted by a Gaussian centered on the interest point with a variance proportional to the interest point scale. The histogram is also smoothed using the approach in the vlfeat implementation; multiple smoothings using a [1/3,1/3,1/3] kernel in a circular manner (i.e. it is assumed the histogram wraps around).

Author:

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

Field Summary

Fields

Modifier and Type	Field and Description
static int	DEFAULT_NUM_BINS Default number of orientation histogram bins; Lowe's IJCV paper (p.13) suggests 36 bins.
static float	DEFAULT_SAMPLING_SIZE Default value for the size of the sampling window relative to the sampling scale.
static float	DEFAULT_SCALING Default value for weighting the scaling Gaussian relative to the keypoint scale.
static int	DEFAULT_SMOOTHING_ITERATIONS Default value for the number of iterations of the smoothing filter.
protected int	numBins
protected float	samplingSize
protected float	scaling
protected int	smoothingIterations

Constructor Summary

Constructors

Constructor and Description
OrientationHistogramExtractor() Default constructor.
OrientationHistogramExtractor(int numBins, float scaling, int smoothingIterations, float samplingSize) Construct with the given parameter values.

Method Summary

Modifier and Type	Method and Description
protected void	circularSmooth(float[] buffer) Smooth the values in a circular buffer with a (1/3)[1,1,1] kernel.
FloatFV[]	extractFeature(GradientScaleSpaceImageExtractorProperties<FImage> props) Extract the orientation histogram given the properties.
float[]	extractFeatureRaw(GradientScaleSpaceImageExtractorProperties<FImage> properties) Extract the orientation histogram given the properties.

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_NUM_BINS

public static int DEFAULT_NUM_BINS

Default number of orientation histogram bins; Lowe's IJCV paper (p.13) suggests 36 bins.

DEFAULT_SCALING

public static float DEFAULT_SCALING

Default value for weighting the scaling Gaussian relative to the keypoint scale. Lowe's IJCV paper (p.13) suggests 1.5.

DEFAULT_SMOOTHING_ITERATIONS

public static int DEFAULT_SMOOTHING_ITERATIONS

Default value for the number of iterations of the smoothing filter. The vlfeat SIFT implementation uses 6.

DEFAULT_SAMPLING_SIZE

public static float DEFAULT_SAMPLING_SIZE

Default value for the size of the sampling window relative to the sampling scale. Lowe's ICCV paper suggests 3;

numBins

protected int numBins

scaling

protected float scaling

smoothingIterations

protected int smoothingIterations

samplingSize

protected float samplingSize

Constructor Detail

OrientationHistogramExtractor

public OrientationHistogramExtractor()

Default constructor.

OrientationHistogramExtractor

public OrientationHistogramExtractor(int numBins,
                                     float scaling,
                                     int smoothingIterations,
                                     float samplingSize)

Construct with the given parameter values.

Parameters:

numBins - number of orientation histogram bins

scaling - weighting for the scaling Gaussian relative to the keypoint scale.

smoothingIterations - the number of iterations of the smoothing filter

samplingSize - size of the sampling window relative to the sampling scale.

Method Detail

extractFeature

public FloatFV[] extractFeature(GradientScaleSpaceImageExtractorProperties<FImage> props)

Extract the orientation histogram given the properties. This method caches gradient and orientation maps as it's likely to be called multiple times (at different positions) for the same input image.

Specified by:

extractFeature in interface FeatureVectorExtractor<FloatFV,GradientScaleSpaceImageExtractorProperties<FImage>>

Parameters:

props - Properties describing the interest point in scale space.

Returns:

a FloatFV object representing the orientation histogram.

extractFeatureRaw

public float[] extractFeatureRaw(GradientScaleSpaceImageExtractorProperties<FImage> properties)

Extract the orientation histogram given the properties. This method caches gradient and orientation maps as it's likely to be called multiple times (at different positions) for the same input image.

Parameters:

properties - Properties describing the interest point in scale space.

Returns:

a float array representing the orientation histogram.

circularSmooth

protected void circularSmooth(float[] buffer)

Smooth the values in a circular buffer with a (1/3)[1,1,1] kernel.

Parameters:

buffer - buffer to smooth