lti::axOGDFeature Class Reference

OGD based texture feature. More...

#include <ltiAxOGDFeature.h>

Inheritance diagram for lti::axOGDFeature:

[legend]

Collaboration diagram for lti::axOGDFeature:

[legend]

List of all members.

Classes
class	parameters
	The parameters for the class axOGDFeature. More...
Public Member Functions
	axOGDFeature ()
	axOGDFeature (const axOGDFeature &other)
virtual	~axOGDFeature ()
virtual const char *	getTypeName () const
virtual bool	use (const channel &src)
virtual bool	use (const channel &center, const channel &surround)
virtual bool	apply (const point &pos, dvector &dest) const
bool	apply (const channel &src, dvector &dest) const
bool	apply (const image &src, dvector &dest) const
bool	apply (const channel &src, channel &a0, channel &a1, channel &theta1) const
bool	apply (const channel &src, channel &a0, channel &a1, channel &a2, channel &theta1, channel &theta2) const
bool	apply (const channel &center, const channel &sround, dvector &dest) const
bool	apply (const channel &src, const std::list< location > &locs, std::list< dvector > &dest)
bool	apply (const image &src, const std::list< location > &locs, std::list< dvector > &dest)
bool	apply (const channel &center, const channel &surround, const std::list< location > &locs, std::list< dvector > &dest)
axOGDFeature &	copy (const axOGDFeature &other)
virtual functor *	clone () const
const parameters &	getParameters () const
Protected Attributes
std::vector< channel >	theChannels
int	ogdOrder
int	numLevels

Detailed Description

OGD based texture feature.

Functor for texture feature extraction that uses the steerability property of the oriented gaussian derivatives (OGD) to generate rotation invariant feature vectors.

The description of this feature can be found on Alvarado, P., Doerfler, P. and Wickel, J. "Axon2 - A visual object recognition system for non-rigid objects" SPPRA July, 2001 (http://www.techinfo.rwth-aachen.de/Veroeffentlichungen/V003_2001.pdf)

The basic idea is to compute the "energy" of a region (a location for local features or the whole image for global features) as a steerable function.

This energy is computed for different "power" channels, which are the result of convolving the images with OGD filters of a specific order.

In some sense, the first order OGD compute some "edge" energy, while second order OGD compute some "line" energy.

Constructor & Destructor Documentation

lti::axOGDFeature::axOGDFeature ( )

Default constructor.

lti::axOGDFeature::axOGDFeature ( const axOGDFeature & other )

Copy constructor.

Parameters:

other

the object to be copied

virtual lti::axOGDFeature::~axOGDFeature ( ) [virtual]

Destructor.

Member Function Documentation

bool lti::axOGDFeature::apply	(	const channel &	center,
		const channel &	surround,
		const std::list< location > &	locs,
		std::list< dvector > &	dest
	)

Extracts the opponent color OGD features for each given location.

The radius of each location is used to determine which scale should be used in the generation of each feature vector.

Parameters:

	center	original channel considered as "center"
	surround	channel considered as surround. This must have the same size as center.
	locs	the locations that need to be analyzed.
	dest	the list of feature vectors. This list will have the same size as locs. The feature vector at position p will correspond to the location at position p.

Returns:: true if successful, of false otherwise.

bool lti::axOGDFeature::apply	(	const image &	src,
		const std::list< location > &	locs,
		std::list< dvector > &	dest
	)

Extracts the OGD features for each given location.

The image will be splitted in its red, green and blue components, and the feature vectors will be concatenated. The radius of each location is used to determine which scale should be used in the generation of each feature vector.

Parameters:

	src	original channel to be analyzed.
	locs	the locations that need to be analyzed.
	dest	the list of feature vectors. This list will have the same size as locs. The feature vector at position p will correspond to the location at position p.

Returns:: true if successful, of false otherwise.

bool lti::axOGDFeature::apply	(	const channel &	src,
		const std::list< location > &	locs,
		std::list< dvector > &	dest
	)

Extracts the OGD features for each given location.

The radius of each location is used to determine which scale should be used in the generation of each feature vector.

Parameters:

	src	original channel to be analyzed.
	locs	the locations that need to be analyzed.
	dest	the list of feature vectors. This list will have the same size as locs. The feature vector at position p will correspond to the location at position p.

Returns:: true if successful, of false otherwise.

bool lti::axOGDFeature::apply	(	const channel &	center,
		const channel &	sround,
		dvector &	dest
	)			const

Extracts the OGD rotation invariant opponent color feature from the given color channels.

Center and Surround must have the same sizes.

Parameters:

	center	the center channel.
	sround	the surround channel.
	dest	the feature vector

Returns:: true if apply successful or false otherwise.

bool lti::axOGDFeature::apply	(	const channel &	src,
		channel &	a0,
		channel &	a1,
		channel &	a2,
		channel &	theta1,
		channel &	theta2
	)			const

Extracts the second order OGD coefficients for the given windows size and variance (if gaussian).

The "levels" parameter will be ignored, and only the first level will be calculated. This method is used by other feature extractors

To obtain the angle othogonal to the edges, you need to divide theta1 by 2. You can use a1 and theta1/2 for the lti::orientationFeature.

bool lti::axOGDFeature::apply	(	const channel &	src,
		channel &	a0,
		channel &	a1,
		channel &	theta1
	)			const

Extracts the first order OGD coefficients for the given window size and variance (if gaussian).

The "levels" parameter will be ignored, and only the first level will be calculated. This method is used by other feature extractors.

To obtain the angle othogonal to the edges, you need to divide theta1 by 2. This can automatically be done if you specify it in the parameters. You can use a1 and theta1/2 for the lti::orientationFeature.

bool lti::axOGDFeature::apply	(	const image &	src,
		dvector &	dest
	)			const `[virtual]`

Extract the OGD features from the red, green and blue components of the given image and concatenate the results.

Parameters:

	src	image with the source data.
	dest	the feature vector

Returns:: true if apply successful or false otherwise.

Reimplemented from lti::globalFeatureExtractor.

bool lti::axOGDFeature::apply	(	const channel &	src,
		dvector &	dest
	)			const `[virtual]`

Extracts the OGD rotation invariant feature from the given gray valued image (a channel).

Parameters:

	src	channel with the source data.
	dest	the feature vector

Returns:: true if apply successful or false otherwise.

Reimplemented from lti::globalFeatureExtractor.

virtual bool lti::axOGDFeature::apply	(	const point &	pos,
		dvector &	dest
	)			const `[virtual]`

Extract the local feature at the position pos and leave the result in the given vector.

It depends on the last used "use" member, if the feature vector is an "opponent-color" feature or a monochromatic feature.

Parameters:

	pos	the position at the given image.
	dest	the feature vector.

Returns:: true if sucessful, false otherwise

virtual functor* lti::axOGDFeature::clone ( ) const [virtual]

Returns a pointer to a clone of this functor.

Reimplemented from lti::globalFeatureExtractor.

axOGDFeature& lti::axOGDFeature::copy ( const axOGDFeature & other )

Copy data of "other" functor.

Parameters:

other

the functor to be copied

Returns:: a reference to this functor object

Reimplemented from lti::globalFeatureExtractor.

const parameters& lti::axOGDFeature::getParameters ( ) const

Returns used parameters.

Reimplemented from lti::globalFeatureExtractor.

virtual const char* lti::axOGDFeature::getTypeName ( ) const [virtual]

Returns the name of this type ("axOGDFeature").

Reimplemented from lti::globalFeatureExtractor.

virtual bool lti::axOGDFeature::use	(	const channel &	center,
		const channel &	surround
	)			`[virtual]`

Indicates which channels to use in the local feature extraction.

Some precalculation will be done, and using the apply(const point&, dvector&) member, the feature vector of a specific position can be computed in a faster way.

After calling this method the corresponding opponent-color features will be extracted.

virtual bool lti::axOGDFeature::use ( const channel & src ) [virtual]

Indicates which channel to use in the local feature extraction.

Some precalculation will be done, and using the apply(const point&, dvector&) member, the feature vector of a specific position can be computed in a faster way.

After calling this method the corresponding monochromatic features will be extracted.

Member Data Documentation

int lti::axOGDFeature::numLevels [protected]

The number of levels included in theChannels.

int lti::axOGDFeature::ogdOrder [protected]

The degree of the OGD used to extract the channels.

std::vector<channel> lti::axOGDFeature::theChannels [protected]

This channels contain the term a0-a2 and theta2.

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

ltiAxOGDFeature.h

lti::axOGDFeature Class Reference

Classes

Public Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation