boofcv.alg.feature.detect.interest

Class SiftImageScaleSpace

java.lang.Object

boofcv.alg.feature.detect.interest.SiftImageScaleSpace

public class SiftImageScaleSpace
extends java.lang.Object

Constructs the scale-space in which SIFT detects features. An octave contains a set of scales. Each octave is half the width/height of the previous octave. Scales are computed inside an octave by applying Gaussian blur. See SIFT paper for the details.

OCTAVE SCALES: The scales sampled are different from the SIFT paper. In each octave the scales are sampled across a linear function. s(i) = sigma*i. In the paper scales are sampled using an exponential function s(i) = sigma*s(i-1)

When computing the next octave in the sequence it is seeded with the image from the second scale in the previous octave. The first octave is seeded with the input image or the input image scaled.

Field Summary

Fields

Modifier and Type	Field and Description
protected int	actualOctaves
protected boofcv.struct.image.ImageFloat32[]	derivX
protected boofcv.struct.image.ImageFloat32[]	derivY
protected boofcv.struct.image.ImageFloat32[]	dog
protected double[]	layerSigma
protected int	numOctaves
protected int	numScales
protected double[]	pixelScale
protected boofcv.struct.image.ImageFloat32[]	scale
protected float	sigma
protected boofcv.struct.image.ImageFloat32	storage

Constructor Summary

Constructors

Constructor and Description
SiftImageScaleSpace(float blurSigma, int numScales, int numOctaves, boolean doubleInputImage) Configures the scale-space.

Method Summary

Methods

Modifier and Type	Method and Description
void	computeDerivatives() Computes the image derivative for each layer in the pyramid.
void	computeFeatureIntensity() Compute difference of Gaussian feature intensity across scale space
double	computeScaleSigma(int octave, int scale) Returns the amount of blur which has been applied to the image in total at the specified scale in the current octave
void	constructPyramid(boofcv.struct.image.ImageFloat32 input) Processes the first image and constructs the scale-space pyramid for the first level.
protected static void	downSample(boofcv.struct.image.ImageFloat32 from, boofcv.struct.image.ImageFloat32 to) Down samples an image by copying every other pixel, starting with pixel 1.
boofcv.struct.image.ImageFloat32	getDerivativeX(int index)
boofcv.struct.image.ImageFloat32	getDerivativeY(int index)
int	getNumOctaves()
int	getNumScales()
boofcv.struct.image.ImageFloat32	getPyramidLayer(int index)
double	imageIndexToPixelScale(int imageIndex) Image pixel scale factor for the specific image in the pyramid.
int	scaleToImageIndex(double sigma) Given the scale, return the index of the layer in the scale-space that is the closest match.
protected static void	upSample(boofcv.struct.image.ImageFloat32 from, boofcv.struct.image.ImageFloat32 to) Up-samples the input image.

Methods inherited from class java.lang.Object

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

Field Detail

numOctaves

protected int numOctaves

numScales

protected int numScales

actualOctaves

protected int actualOctaves

dog

protected boofcv.struct.image.ImageFloat32[] dog

scale

protected boofcv.struct.image.ImageFloat32[] scale

derivX

protected boofcv.struct.image.ImageFloat32[] derivX

derivY

protected boofcv.struct.image.ImageFloat32[] derivY

sigma

protected float sigma

pixelScale

protected double[] pixelScale

layerSigma

protected double[] layerSigma

storage

protected boofcv.struct.image.ImageFloat32 storage

Constructor Detail

SiftImageScaleSpace

public SiftImageScaleSpace(float blurSigma,
                   int numScales,
                   int numOctaves,
                   boolean doubleInputImage)

Configures the scale-space.

Parameters:

blurSigma - Amount of blur applied to each scale inside an octaves. Try 1.6

numScales - Number of scales per octaves. Try 5. Must be >= 3

numOctaves - Number of octaves to detect. Try 4

doubleInputImage - Should the input image be doubled? Try false.

Method Detail

constructPyramid

public void constructPyramid(boofcv.struct.image.ImageFloat32 input)

Processes the first image and constructs the scale-space pyramid for the first level.

Parameters:

input - Input image

computeDerivatives

public void computeDerivatives()

Computes the image derivative for each layer in the pyramid.

computeScaleSigma

public double computeScaleSigma(int octave,
                       int scale)

Returns the amount of blur which has been applied to the image in total at the specified scale in the current octave

computeFeatureIntensity

public void computeFeatureIntensity()

Compute difference of Gaussian feature intensity across scale space

downSample

protected static void downSample(boofcv.struct.image.ImageFloat32 from,
              boofcv.struct.image.ImageFloat32 to)

Down samples an image by copying every other pixel, starting with pixel 1.

upSample

protected static void upSample(boofcv.struct.image.ImageFloat32 from,
            boofcv.struct.image.ImageFloat32 to)

Up-samples the input image. Doubling its size.

getNumOctaves

public int getNumOctaves()

getNumScales

public int getNumScales()

getPyramidLayer

public boofcv.struct.image.ImageFloat32 getPyramidLayer(int index)

getDerivativeX

public boofcv.struct.image.ImageFloat32 getDerivativeX(int index)

getDerivativeY

public boofcv.struct.image.ImageFloat32 getDerivativeY(int index)

scaleToImageIndex

public int scaleToImageIndex(double sigma)

Given the scale, return the index of the layer in the scale-space that is the closest match.

imageIndexToPixelScale

public double imageIndexToPixelScale(int imageIndex)

Image pixel scale factor for the specific image in the pyramid.