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

Class GradientFeatureExtractor

java.lang.Object

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

All Implemented Interfaces:

ScaleSpaceFeatureExtractor<OrientedFeatureVector,FImage>, FeatureVectorExtractor<OrientedFeatureVector,ScaleSpaceImageExtractorProperties<FImage>>

public class GradientFeatureExtractor
extends Object
implements ScaleSpaceFeatureExtractor<OrientedFeatureVector,FImage>

Class capable of extracting local descriptors from a circular region in an image defined by its scale and centre. The actual feature extracted is determined by the GradientFeatureProvider that is provided by the GradientFeatureProviderFactory set during construction.

The GradientFeatureExtractor first calculates the dominant orientation of the image patch described by the ScaleSpaceImageExtractorProperties and then iterates over the pixels in an oriented square, centered on the interest point, passing the gradient and magnitude values of the respective pixel to the GradientFeatureProvider.

The size of the sampling square, relative to scale is set by a single parameter, magnification. For some types of feature provider, this number might need to be set based on the internal settings of the provider. For example, with a SIFTFeatureProvider this will probably be set to a constant multiplied by the number of spatial bins of the feature. For SIFT, this constant is typically around 3, so with a standard 4-spatial binned SIFT provider, the magnification factor of the extractor should be about 12.

Author:

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

Field Summary

Fields

Modifier and Type	Field and Description
protected float	magnification The magnification factor determining the size of the sampling region relative to the scale of the interest point.

Constructor Summary

Constructors

Constructor and Description
GradientFeatureExtractor(AbstractDominantOrientationExtractor dominantOrientationExtractor, GradientFeatureProviderFactory factory) Construct with the given orientation extractor and gradient feature provider.
GradientFeatureExtractor(AbstractDominantOrientationExtractor dominantOrientationExtractor, GradientFeatureProviderFactory factory, float magnification) Construct with the given orientation extractor, gradient feature provider and magnification factor determining the size of the sampling region relative to the scale of the interest point.

Method Summary

Modifier and Type	Method and Description
protected OrientedFeatureVector	createFeature(float orientation)
OrientedFeatureVector[]	extractFeature(ScaleSpaceImageExtractorProperties<FImage> properties) Extract FeatureVector(s), possibly using the information given in the properties.
GradientScaleSpaceImageExtractorProperties<FImage>	getCurrentGradientProps(ScaleSpaceImageExtractorProperties<FImage> properties) Get the GradientScaleSpaceImageExtractorProperties for the given properties.

Methods inherited from class java.lang.Object

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

Field Detail

magnification

protected float magnification

The magnification factor determining the size of the sampling region relative to the scale of the interest point.

Constructor Detail

GradientFeatureExtractor

public GradientFeatureExtractor(AbstractDominantOrientationExtractor dominantOrientationExtractor,
                                GradientFeatureProviderFactory factory)

Construct with the given orientation extractor and gradient feature provider. The default magnification factor of 12 is used.

Parameters:

dominantOrientationExtractor - the orientation extractor

factory - the gradient feature provider

GradientFeatureExtractor

public GradientFeatureExtractor(AbstractDominantOrientationExtractor dominantOrientationExtractor,
                                GradientFeatureProviderFactory factory,
                                float magnification)

Construct with the given orientation extractor, gradient feature provider and magnification factor determining the size of the sampling region relative to the scale of the interest point.

Parameters:

dominantOrientationExtractor - the orientation extractor

factory - the gradient feature provider

magnification - the magnification factor.

Method Detail

extractFeature

public OrientedFeatureVector[] extractFeature(ScaleSpaceImageExtractorProperties<FImage> properties)

Description copied from interface: FeatureVectorExtractor

Extract FeatureVector(s), possibly using the information given in the properties.

Specified by:

extractFeature in interface FeatureVectorExtractor<OrientedFeatureVector,ScaleSpaceImageExtractorProperties<FImage>>

Parameters:

properties - Properties describing how the features should be extracted.

Returns:

An array of feature vectors.

getCurrentGradientProps

public GradientScaleSpaceImageExtractorProperties<FImage> getCurrentGradientProps(ScaleSpaceImageExtractorProperties<FImage> properties)

Get the GradientScaleSpaceImageExtractorProperties for the given properties. The returned properties are the same as the input properties, but with the gradient images added. For efficiency, this method always returns the same cached GradientScaleSpaceImageExtractorProperties, and internally updates this as necessary. The gradient images are only recalculated when the input image from the input properties is different to the cached one.

Parameters:

properties - input properties

Returns:

cached GradientScaleSpaceImageExtractorProperties

createFeature

protected OrientedFeatureVector createFeature(float orientation)