---

fl::InterestDOG Class Reference

#include <interest.h>

Inheritance diagram for fl::InterestDOG:

List of all members.

Public Methods
	InterestDOG (float firstScale=1.6f, float lastScale=INFINITY, int extraSteps=3)
	extraSteps gives the number of sub-levels between octaves.
virtual void	run (const Image &image, std::multiset< PointInterest > &result)
bool	isLocalMax (float value, ImageOf< float > &dog, int x, int y)
bool	notOnEdge (ImageOf< float > &dog, int x, int y)
float	fitQuadratic (std::vector< ImageOf< float > > &dogs, int s, int x, int y, Vector< float > &result)
Public Attributes
std::vector< Image >	pyramid
	A series of blurred/resampled images. pyramid[0] contains the original image in GrayFloat (or GrayDouble if that was the original format). The coordinates in pyramid[i] are (x/s,y/s) where (x,y) are the coordinates in pyramid[0] and s = pyramid[0].width / pyramid[i].width.
std::vector< float >	scales
	Scale levels associated with the entries in pyramid.
float	firstScale
float	lastScale
int	steps
	Number scale steps between octaves.
int	crop
	Number of pixels from border to ignore maxima.
bool	storePyramid
	Indicates that blurred/resampled images should be stored rather than thrown away.
float	thresholdEdge
	Gives smallest permissible ratio of det(H) / trace(H)^2, where H is the Hessian of the DoG function on intensity.
float	thresholdPeak
	Minimum permissible strength of DoG function at a local maximum.

---

Detailed Description

Implements David Lowe's scale pyramid approach to finding difference of Gaussian extrema. The shape of a difference-of-Gaussian kernel is very similar to a Laplacian of Gaussian.

The member "pyramid" stores blurred images that could be used to generate gradient info for use DescriptorSIFT.

---

Constructor & Destructor Documentation

InterestDOG::InterestDOG (  float    firstScale = 1.6f, float    lastScale = INFINITY, int    extraSteps = 3 )

extraSteps gives the number of sub-levels between octaves.

---

Member Function Documentation

float InterestDOG::fitQuadratic (  std::vector< ImageOf< float > > &    dogs, int    s, int    x, int    y, Vector< float > &    result )

bool InterestDOG::isLocalMax (  float    value, ImageOf< float > &    dog, int    x, int    y )  [inline]

bool InterestDOG::notOnEdge (  ImageOf< float > &    dog, int    x, int    y )  [inline]

void InterestDOG::run (  const Image &    image, std::multiset< PointInterest > &    result )  [virtual]

Implements fl::InterestOperator.

---

Member Data Documentation

int fl::InterestDOG::crop

Number of pixels from border to ignore maxima.

float fl::InterestDOG::firstScale

float fl::InterestDOG::lastScale

std::vector<Image> fl::InterestDOG::pyramid

A series of blurred/resampled images. pyramid[0] contains the original image in GrayFloat (or GrayDouble if that was the original format). The coordinates in pyramid[i] are (x/s,y/s) where (x,y) are the coordinates in pyramid[0] and s = pyramid[0].width / pyramid[i].width.

std::vector<float> fl::InterestDOG::scales

Scale levels associated with the entries in pyramid.

int fl::InterestDOG::steps

Number scale steps between octaves.

bool fl::InterestDOG::storePyramid

Indicates that blurred/resampled images should be stored rather than thrown away.

float fl::InterestDOG::thresholdEdge

Gives smallest permissible ratio of det(H) / trace(H)^2, where H is the Hessian of the DoG function on intensity.

float fl::InterestDOG::thresholdPeak

Minimum permissible strength of DoG function at a local maximum.

---

The documentation for this class was generated from the following files:

• interest.h
• InterestDOG.cc

---