lti::kullbackContrast Class Reference

Calculate the Kullback-Constrast for a channel, as described in: Jagersand, Martin. More...

#include <ltiKullbackContrast.h>

Inheritance diagram for lti::kullbackContrast:

[legend]

Collaboration diagram for lti::kullbackContrast:

[legend]

List of all members.

Classes
class	parameters
	the parameters for the class kullbackContrast More...
Public Member Functions
	kullbackContrast ()
	kullbackContrast (const kullbackContrast &other)
virtual	~kullbackContrast ()
virtual const char *	getTypeName () const
vector< float > &	apply (const channel &src, vector< float > &dest) const
vector< float > &	apply (const channel &src, vector< float > &dest, gaussianPyramid< channel > &pyr, gaussianPyramid< channel > &pyr2) const
float &	apply (const channel &q, const channel &p, float &dest) const
float &	apply (const vector< float > &q, const vector< float > &p, float &dest) const
kullbackContrast &	copy (const kullbackContrast &other)
virtual functor *	clone () const
const parameters &	getParameters () const
Protected Member Functions
void	makeHistogram (const channel &src, vector< double > &hist) const

Detailed Description

Calculate the Kullback-Constrast for a channel, as described in: Jagersand, Martin.

Saliency Maps and Attention Selection in Scale and Spatial Coordinates: An Information Theoretic Approach, 5th Int. Conf. on Computer Vision, 1995, pp. 195-202

The given channel will be analized at different resolutions (given through the parameters). The output vector contains the constrast for each size, where the evaluated size corresponds to 2^(vector index).

Note that if one of the distributions contains a zero (p(x_i) = 0) and q(x_i) != 0, the Kullback constrast will be infinite!

With the parameter useHistogram an approximation for the probability distribution of the intensity of a channel will be done in form of a histogram. Please note that this will option is ignored when comparing vectors.

Constructor & Destructor Documentation

lti::kullbackContrast::kullbackContrast ( )

default constructor

lti::kullbackContrast::kullbackContrast ( const kullbackContrast & other )

copy constructor

Parameters:

other

the object to be copied

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

destructor

Member Function Documentation

float& lti::kullbackContrast::apply	(	const vector< float > &	q,
		const vector< float > &	p,
		float &	dest
	)			const

calculates the Kullback contrast between q and p, defined as

$K[Q,P] = \int_x p(x) \log \frac{p(x)}{q(x)} dx$

operates on a copy of the given parameters.

Parameters:

	q	vector with the source data q(x).
	p	vector with the source data p(x).
	dest	vector<float> where the result will be left.

Returns:: a reference to the dest.

float& lti::kullbackContrast::apply	(	const channel &	q,
		const channel &	p,
		float &	dest
	)			const

calculates the Kullback contrast between q and p, defined as

$K[Q,P] = \int_x p(x) \log \frac{p(x)}{q(x)} dx$

operates on a copy of the given parameters. The input channels do not need to be "normalized". This will be done here, in an efficient way.

Parameters:

	q	channel with the source data q(x).
	p	channel with the source data p(x).
	dest	channel where the result will be left.

Returns:: a reference to the dest.

vector<float>& lti::kullbackContrast::apply	(	const channel &	src,
		vector< float > &	dest,
		gaussianPyramid< channel > &	pyr,
		gaussianPyramid< channel > &	pyr2
	)			const

The Kullback contrast of the channel src will be calculated as the Kullback contrast between two adjacent resolutions of the gaussian pyramid for that channel, where the pixel value normalized with the sum of all pixels will be considered as a probability p(x,y).

(for more details see the paper: Jagersand, Martin. "Saliency Maps and Attention Selection in Scale and Spatial Coordinates: An Information Theoretic Approach", 5th Int. Conf. on Computer Vision, 1995, pp. 195-202 operates on a copy of the given parameters.

Parameters:

	src	channel with the source data.
	dest	channel where the result will be left. The i-th entry in the vector will contain the Kullback contrast for objects with a diameter of 2^i pixels.
	pyr	the gaussian pyramid of the given channel
	pyr2	the upsampled gaussian pyramid (pyr2.size() == pyr.size()-1)

Returns:: a reference to the dest.

vector<float>& lti::kullbackContrast::apply	(	const channel &	src,
		vector< float > &	dest
	)			const

The Kullback contrast of the channel src will be calculated as the Kullback contrast between two adjacent resolutions of the gaussian pyramid for that channel, where the pixel value normalized with the sum of all pixels will be considered as a probability p(x,y).

(for more details see the paper: Jagersand, Martin. "Saliency Maps and Attention Selection in Scale and Spatial Coordinates: An Information Theoretic Approach", 5th Int. Conf. on Computer Vision, 1995, pp. 195-202 operates on a copy of the given parameters.

Parameters:

	src	channel with the source data.
	dest	channel where the result will be left. The i-th entry in the vector will contain the Kullback contrast for objects with a diameter of 2^i pixels.