latest version v1.9 - last update 24 Nov 2005

lti::meanShiftSegmentation Class Reference

This is the implementation of the mean shift segmentation algorithm. More...

#include <ltiMeanShiftSegmentation.h>

Inheritance diagram for lti::meanShiftSegmentation:

[legend]Collaboration diagram for lti::meanShiftSegmentation:

[legend]List of all members.

Public Member Functions
	meanShiftSegmentation ()
	meanShiftSegmentation (const parameters &par)
	meanShiftSegmentation (const meanShiftSegmentation &other)
virtual	~meanShiftSegmentation ()
virtual const char *	getTypeName () const
bool	apply (const lti::image &src, lti::image &dest)
bool	apply (const lti::image &src, lti::imatrix &dest)
bool	apply (const lti::image &src, lti::imatrix &labels, lti::palette &colorMap)
bool	apply (const lti::channel8 &chnl1, const lti::channel8 &chnl2, const lti::channel8 &chnl3, lti::imatrix &dest)
bool	apply (const lti::image &src, lti::image &destFiltered, lti::image &destSegmented, lti::imatrix &destLabels, lti::palette &destColorMap)
bool	filter (const lti::image &src, lti::image &dest)
bool	filter (lti::image &srcdest)
meanShiftSegmentation &	copy (const meanShiftSegmentation &other)
virtual functor *	clone () const
const parameters &	getParameters () const
Classes
struct	luvPixel
	like rgbPixel for RGB-ColorSpace, is luvPixel for LUV-ColorSpace More...
class	parameters
	the parameters for the class meanShiftSegmentation More...

---

Detailed Description

This is the implementation of the mean shift segmentation algorithm.

It is described in

D. Comaniciu, P. Meer, "Mean Shift: A Robust Approach toward Feature Space Analysis" appeared in IEEE Trans. Pattern Analysis Machine Intell. Vol. 24, No. 5, 603-619, 2002 (original link here )

http://www.caip.rutgers.edu/~comanici/Papers/MsRobustApproach.pdf

The code here is based on the code of the original authors, modified extensively for the use with the LTI-Lib. It has changed a lot since then to provide more functionality and configuration options.

There are two implementations available:

The classic implementation is selected setting the parameter parameters::classicAlgorithm to true. It is based on an old code example published by Comaniciu and colleagues a few years ago. This implementation was not as good as the newer version, but was already available in the LTI-Lib. Since several applications were adapted to use that mean-shift version, the default behaviour of this functors has not been changed.

The new implementation is an adaption of Comaniciu and Meer's EDISON code (http://www.caip.rutgers.edu/riul/research/robust.html). It can be selected setting parameters::classicAlgorithm to false. Decisive for the result of the new implementation is the choice of the parameters parameters::sigmaR, parameters::sigmaS, parameters::maxNeighbourColorDistance, and the parameters::speedup.

Warning:

This functor will be completely reviewed soon. Its interface may change a little bit to improve its usability. It has now the problem that the parameters for the classic version and the ones for the new algorithmic implementation are different and they shouldn't.

Todo:

The interface for classic and new algorithms has to be unified and simplified.

There are also several efficiency aspects to be reviewed.

---

Constructor & Destructor Documentation

lti::meanShiftSegmentation::meanShiftSegmentation (   )

default constructor

lti::meanShiftSegmentation::meanShiftSegmentation (  const parameters &  par  )

default constructor

lti::meanShiftSegmentation::meanShiftSegmentation (  const meanShiftSegmentation &  other  )

copy constructor Parameters: other  the object to be copied

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

destructor

---

Member Function Documentation

bool lti::meanShiftSegmentation::apply (  const lti::image &  src, lti::image &  destFiltered, lti::image &  destSegmented, lti::imatrix &  destLabels, lti::palette &  destColorMap )

Apply the mean shift segmentation algorithm to the given image. This image will be internally transformed into the Luv color space, as proposed by Comaniciu et.al. in their original paper. Parameters: src  image with the source data. destFiltered  the meanshift filtered image (not available for classic algorithm) destSegmented  the segmented image destLabels  label associating each pixel to a mode and thus a color to each pixel destColorMap  the colors of the found modes Returns: true if apply successful or false otherwise.

bool lti::meanShiftSegmentation::apply (  const lti::channel8 &  chnl1, const lti::channel8 &  chnl2, const lti::channel8 &  chnl3, lti::imatrix &  dest )

Apply the mean shift segmentation algorithm to a color image, which has been already split into three channels. The algorithm will work with these channels without converting them into the Luv color space. This allows the use of the mean shift segmentation in other color spaces than Luv. Parameters: chnl1  first channel of the source color image. chnl2  second channel of the source color image. chnl3  third channel of the source color image. dest  imatrix where the result (region-labels) will be left. Returns: true if apply successful or false otherwise.

bool lti::meanShiftSegmentation::apply (  const lti::image &  src, lti::imatrix &  labels, lti::palette &  colorMap )

Apply the mean shift segmentation algorithm to the given image. This image will be internally transformed into the Luv color space, as proposed by Comaniciu et.al. in their original paper. Parameters: src  image with the source data. labels  imatrix where the result (region-labels) will be left. colorMap  the color palette used Returns: true if apply successful or false otherwise.

bool lti::meanShiftSegmentation::apply (  const lti::image &  src, lti::imatrix &  dest )

Apply the mean shift segmentation algorithm to the given image. This image will be internally transformed into the Luv color space, as proposed by Comaniciu et.al. in their original paper. Parameters: src  image with the source data. dest  imatrix where the result (region-labels) will be left. Returns: true if apply successful or false otherwise.

bool lti::meanShiftSegmentation::apply (  const lti::image &  src, lti::image &  dest )

Apply the mean shift segmentation algorithm to the given image. This image will be internally transformed into the Luv color space, as proposed by Comaniciu et.al. in their original paper. Parameters: src  image with the source data. dest  image where the result (segmented image) will be left. Returns: true if apply successful or false otherwise.

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

returns a pointer to a clone of this functor. Reimplemented from lti::segmentation.

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

copy data of "other" functor. Parameters: other  the functor to be copied Returns: a reference to this functor object

bool lti::meanShiftSegmentation::filter (  lti::image &  srcdest  )

filter the image with meanshift algorithm Parameters: srcdest  the image to be filtered Returns: true if apply successful or false otherwise.

bool lti::meanShiftSegmentation::filter (  const lti::image &  src, lti::image &  dest )

filter the image with meanshift algorithm Parameters: src  the image to be filtered dest  the result will be left here Returns: true if apply successful or false otherwise.

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

returns used parameters Reimplemented from lti::functor.

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

returns the name of this type ("meanShiftSegmentation") Reimplemented from lti::segmentation.

---

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

• ltiMeanShiftSegmentation.h

---