w2img::GaussianMixtureModel Class Reference

List of all members.

---

Detailed Description

Reports parameters of the best-fit Gaussian Mixture Model to a set of data:.

        P(x,y) =  wi N(x,y given xi,yi,sxi,syi)
     

where P is the final density function; xi,yi,sxi,syi are the mean and standard deviation of a Gaussian; N(x,y) is the value of the Gaussian at x,y; wi is the weight of this Gaussian

Author:

Lakshman

Version:

Id

w2img_GaussianMixtureModel.h,v 1.15 2010/06/23 21:33:50 lakshman Exp

Static Public Member Functions
static void	init_gmm (const std::vector< float > &x, const std::vector< float > &y, int n_models, std::vector< GaussianComponent > &mixture)
	simply slices the input and initializes a GMM fit to each slice.
static void	init_gmm (const std::vector< float > &x, const std::vector< float > &y, const std::vector< int > &startindex, int desired_n_models, std::vector< GaussianComponent > &mixture)
	Uses user-defined slices and initializes a GMM fit.
static float	tune_gmm_em (const std::vector< float > &x, const std::vector< float > &y, std::vector< GaussianComponent > &mixture, const float min_improvement=0.05, const int max_iter=10)
	Tunes an initialized GMM using the Expectation-Minimization algorithm.
static float	fit (const std::vector< float > &x, const std::vector< float > &y, std::vector< GaussianComponent > &mixture, int max_n_models=-1)
	Uses the Bayesian Information Criterion to come up with a reasonable number for number of models.
Classes
class	GaussianComponent

---

Member Function Documentation

static float w2img::GaussianMixtureModel::fit	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		std::vector< GaussianComponent > &	mixture,
		int	max_n_models = -1
	)			[static]

Uses the Bayesian Information Criterion to come up with a reasonable number for number of models.

Returns:

log(likelihood) of the complete fit.

If max_n_models is not specified, x.size()/5 (no reduction) is used.

static void w2img::GaussianMixtureModel::init_gmm	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		const std::vector< int > &	startindex,
		int	desired_n_models,
		std::vector< GaussianComponent > &	mixture
	)			[static]

Uses user-defined slices and initializes a GMM fit.

The startindex provides the starting index for each slice. Slices will be combined/split to match the desired number of models.

Returns:

log(likelihood) of the complete fit.

static void w2img::GaussianMixtureModel::init_gmm	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		int	n_models,
		std::vector< GaussianComponent > &	mixture
	)			[static]

simply slices the input and initializes a GMM fit to each slice.

Returns:

log(likelihood) of the complete fit.

static float w2img::GaussianMixtureModel::tune_gmm_em	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		std::vector< GaussianComponent > &	mixture,
		const float	min_improvement = 0.05,
		const int	max_iter = 10
	)			[static]

Tunes an initialized GMM using the Expectation-Minimization algorithm.

Returns:

log(likelihood) of the complete fit.

Parameters:

	min_improvement	If improvement in log(likelihood) is less than this ratio (0.05 indicates 5%), EM process is stopped.
	max_iter	Maximum number of EM iterations

---