utilities.boofcv_extensions
Class DescribePointSiftNormalization

java.lang.Object
  utilities.boofcv_extensions.DescribePointSiftNormalization

public class DescribePointSiftNormalization
extends java.lang.Object

Detects SIFT features inside the provided SIFT scale-space. SIFT features work by sampling the image in a grid. For each grid element, a histogram of the gradient's orientation is computed. From this histogram the descriptor is computed. See [1] for the details and below for algorithmic changes.

DESCRIPTOR INTERPOLATION: Instead of using trilinear interpolation a Gaussian weight is used instead. Both methods were tried and Gaussian weight produced slightly better results. Same results had been found by others with regard to SURF descriptors.

[1] Lowe, D. "Distinctive image features from scale-invariant keypoints". International Journal of Computer Vision, 60, 2 (2004), pp.91--110.

Author:

Peter Abeles

Field Summary

private double	angleStep
private boofcv.struct.image.ImageFloat32	derivX
private boofcv.struct.image.ImageFloat32	derivY
private double[]	gridWeights
private int	gridWidth
private double[][]	histograms
private boofcv.struct.image.ImageFloat32	image
private int	numHistBins
private int	numSamples
private double	sigmaToRadius
private boofcv.alg.feature.detect.interest.SiftImageScaleSpace	ss

Constructor Summary

DescribePointSiftNormalization(int gridWidth, int numSamples, int numHistBins, double weightSigma, double sigmaToRadius)
Configures detector

Method Summary

private void	addToHistograms(int gridX, int gridY, double locX, double locY, double gradX, double gradY)
private void	computeDescriptor(boofcv.struct.feature.SurfFeature desc)
private void	constructHistograms(double c_x, double c_y, double scale, double orientation)
int	getDescriptorLength()
void	process(double c_x, double c_y, double scale, double orientation, int imageIndex, double pixelScale, boofcv.struct.feature.SurfFeature desc) Compute the descriptor with information on which level in the scale-space to use.
void	process(double c_x, double c_y, double scale, double orientation, boofcv.struct.feature.SurfFeature desc)
void	setScaleSpace(boofcv.alg.feature.detect.interest.SiftImageScaleSpace ss)

Methods inherited from class java.lang.Object

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

Field Detail

ss

private boofcv.alg.feature.detect.interest.SiftImageScaleSpace ss

sigmaToRadius

private double sigmaToRadius

gridWidth

private int gridWidth

numSamples

private int numSamples

numHistBins

private int numHistBins

angleStep

private double angleStep

image

private boofcv.struct.image.ImageFloat32 image

derivX

private boofcv.struct.image.ImageFloat32 derivX

derivY

private boofcv.struct.image.ImageFloat32 derivY

histograms

private double[][] histograms

gridWeights

private double[] gridWeights

Constructor Detail

DescribePointSiftNormalization

public DescribePointSiftNormalization(int gridWidth,
                                      int numSamples,
                                      int numHistBins,
                                      double weightSigma,
                                      double sigmaToRadius)

Configures detector

Parameters:

gridWidth - Number of grid elements along a side. Typically 4

numSamples - Number of samples along a grid. Typically 8

numHistBins - Number of bins in the orientation histogram. Typically 8

weightSigma - Adjusts descriptor element's weighting from center. Typically 0.5

sigmaToRadius - Conversation from scale space to pixels. Typically 2.5

Method Detail

setScaleSpace

public void setScaleSpace(boofcv.alg.feature.detect.interest.SiftImageScaleSpace ss)

process

public void process(double c_x,
                    double c_y,
                    double scale,
                    double orientation,
                    boofcv.struct.feature.SurfFeature desc)

process

public void process(double c_x,
                    double c_y,
                    double scale,
                    double orientation,
                    int imageIndex,
                    double pixelScale,
                    boofcv.struct.feature.SurfFeature desc)

Compute the descriptor with information on which level in the scale-space to use.

constructHistograms

private void constructHistograms(double c_x,
                                 double c_y,
                                 double scale,
                                 double orientation)

addToHistograms

private void addToHistograms(int gridX,
                             int gridY,
                             double locX,
                             double locY,
                             double gradX,
                             double gradY)

computeDescriptor

private void computeDescriptor(boofcv.struct.feature.SurfFeature desc)

getDescriptorLength

public int getDescriptorLength()