lti::coilBackgroundSegmentation Class Reference

The coilBackgroundSegmentation functor tries to detect which pixels belong to the background and which not. More...

#include <ltiCoilBackgroundSegmentation.h>

Inheritance diagram for lti::coilBackgroundSegmentation:

[legend]

Collaboration diagram for lti::coilBackgroundSegmentation:

[legend]

List of all members.

Classes
class	parameters
	the parameters for the class coilBackgroundSegmentation More...
struct	TmeanProbability
	type combining the region with its mean color, probability and relation to background and object More...
Public Member Functions
	coilBackgroundSegmentation ()
	coilBackgroundSegmentation (const coilBackgroundSegmentation &other)
virtual	~coilBackgroundSegmentation ()
virtual const char *	getTypeName () const
bool	apply (const image &src, channel8 &result)
coilBackgroundSegmentation &	copy (const coilBackgroundSegmentation &other)
coilBackgroundSegmentation &	operator= (const coilBackgroundSegmentation &other)
virtual functor *	clone () const
const parameters &	getParameters () const
Protected Member Functions
void	deleteWeakLinks (channel8 &ch8) const
bool	getInvertedCovarMatrixForMask (const image &img, channel8 &mask, trgbPixel< double > &mean, dmatrix &covarMatrix) const
bool	checkColorReasonable (const trgbPixel< double > &mean, const dmatrix &covarMatrix, const rgbPixel &xMinusM) const
bool	growOutsideObject (const image &img, channel8 &ch8MaskChannel, sparseMatrix< int > &intSparseMat, std::vector< TmeanProbability > &meanProbabilityList, matrix< int > &newLabelsMatrix, bool &converged) const
bool	growInsideObject (const image &img, channel8 &ch8MaskChannel, sparseMatrix< int > &intSparseMat, std::vector< TmeanProbability > &meanProbabilityList, matrix< int > &newLabelsMatrix, bool &converged) const

---

Detailed Description

The coilBackgroundSegmentation functor tries to detect which pixels belong to the background and which not.

It is quite slow, so it should be used mainly for static applications where the mask has to be generated only once and without any matter of time.

For proper functionality there are some assumtions/requirements:

• The background has to be almost homogeneous. (The functor works best with black background, but this is not necessary.)
• There is a high probability that the border of the image contains only background pixels and the object overlaps the middle of the image.

The algorithm works as follows:

At first the colors are quantized into regions using the k-means algorithm. Then two masks are generated, one representing the object, the other one the background. These masks serve as a basis for a probability estimation which region belongs to object or background. After that the masks grow considering these results, neighbourhoods and covariance matrices. This iteration is repeated "iteration"-times and finally generates a channel containing 0 for background and 255 for the object.

At last some after-treatment is done: Pixels constituting weak links are deleted and if there is more than one object estimated, only the largest one is kept. In order to improve results, the whole algorithm is reiterated up to "maxNumberBestOfN"-times, erasing pixels in the final mask which are not estimated as object every time. In the best case these shrinking lasts until the final mask doesn´t change any more.

See the lti::coilBackgroundSegmentation::parameters for more information.

---

Constructor & Destructor Documentation

lti::coilBackgroundSegmentation::coilBackgroundSegmentation

(

)

default constructor

lti::coilBackgroundSegmentation::coilBackgroundSegmentation

(

const coilBackgroundSegmentation &

other

)

copy constructor

Parameters:

other

the object to be copied

virtual lti::coilBackgroundSegmentation::~coilBackgroundSegmentation

(

)

[virtual]

destructor

---

Member Function Documentation

bool lti::coilBackgroundSegmentation::apply	(	const image &	src,
		channel8 &	result
	)

generates a mask with the value 255 where the object is estimated and 0 for the background.

Parameters:

	src	image with the source data
	result	the resulting mask

Returns:

true if apply successful or false otherwise

bool lti::coilBackgroundSegmentation::checkColorReasonable	(	const trgbPixel< double > &	mean,
		const dmatrix &	covarMatrix,
		const rgbPixel &	xMinusM
	)			const [protected]

checking if a given color is close enough to the covariance matrix

Parameters:

	mean	the center of the covarianve matrix is considered here
	covarMatrix	the covariance matrix to be considered
	xMinusM	pixel with the color to be checked

Returns:

true if succesfull

virtual functor* lti::coilBackgroundSegmentation::clone

(

)

const [virtual]

returns a pointer to a clone of this functor.

Reimplemented from lti::segmentation.

coilBackgroundSegmentation& lti::coilBackgroundSegmentation::copy

(

const coilBackgroundSegmentation &

other

)

copy data of "other" functor.

Parameters:

other

the functor to be copied

Returns:

a reference to this functor object

Reimplemented from lti::segmentation.

void lti::coilBackgroundSegmentation::deleteWeakLinks

(

channel8 &

ch8

)

const [protected]

delete such positions in the mask

    ** **        *****
    ** **        *****
    **x**          x
    ** **        *****
    ** **        *****

Parameters:

ch8

the channel8 to be optimized

bool lti::coilBackgroundSegmentation::getInvertedCovarMatrixForMask	(	const image &	img,
		channel8 &	mask,
		trgbPixel< double > &	mean,
		dmatrix &	covarMatrix
	)			const [protected]

calculating the inverted covariance matrix under consideration of the object/background mask

Parameters:

	img	the image to be considered
	mask	the mask which pixels from img should be used
	mean	the meanvalue (center) of the covarianve matrix is returned here
	covarMatrix	the resulting matrix is returned here

Returns:

true if succesfull

const parameters& lti::coilBackgroundSegmentation::getParameters

(

)

const

returns used parameters

Reimplemented from lti::functor.

virtual const char* lti::coilBackgroundSegmentation::getTypeName

(

)

const [virtual]

returns the name of this type ("coilBackgroundSegmentation")

Reimplemented from lti::segmentation.

bool lti::coilBackgroundSegmentation::growInsideObject	(	const image &	img,
		channel8 &	ch8MaskChannel,
		sparseMatrix< int > &	intSparseMat,
		std::vector< TmeanProbability > &	meanProbabilityList,
		matrix< int > &	newLabelsMatrix,
		bool &	converged
	)			const [protected]

building the object mask under consideration of neighbourhoods, covariance matrix and probabilities

Parameters:

	img	the image to be considered
	ch8MaskChannel	the mask estimating the object
	intSparseMat	matix containing which labels are related
	meanProbabilityList	a list containing mean probability, color and relation towards background and object for each region
	newLabelsMatrix	matrix containing the labelnumber for each pixel
	converged	true if no change took place

Returns:

true if succesfull

bool lti::coilBackgroundSegmentation::growOutsideObject	(	const image &	img,
		channel8 &	ch8MaskChannel,
		sparseMatrix< int > &	intSparseMat,
		std::vector< TmeanProbability > &	meanProbabilityList,
		matrix< int > &	newLabelsMatrix,
		bool &	converged
	)			const [protected]

building the background mask under consideration of neighbourhoods, covariance matrix and probabilities

Parameters:

	img	the image to be considered
	ch8MaskChannel	the mask estimating the background
	intSparseMat	matix containing which regions are neighbours
	meanProbabilityList	a list containing mean probability, color and relation towards background and object for each region
	newLabelsMatrix	matrix containing the labelnumber for each pixel
	converged	true if no change took place

Returns:

true if succesfull

coilBackgroundSegmentation& lti::coilBackgroundSegmentation::operator=

(

const coilBackgroundSegmentation &

other

)

alias for copy member

Parameters:

other

the functor to be copied

Returns:

a reference to this functor object

Reimplemented from lti::functor.

---

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

• ltiCoilBackgroundSegmentation.h