ClassifierSelector Class Reference

#include <ClassifierSelector.h>

Public Member Functions
	ClassifierSelector (int numW, const Size &patchSize, int numB=2)
	ClassifierSelector (int numW, WeakClassifier **weaks, int numB=2)
void	Initialize (const Size &patchSize)
void	Train (const IntegralImage *intImage, const Rect &roi, int target, float importance, bool *errMask)
float	GetError (int index=-1) const
virtual int	SelectBestClassifer (float importance, const bool *errorMask, float *errors)
virtual int	ReplaceWeakestClassifier (float *sumErrors, const Size &patchSize)
virtual void	ReplaceWeakestClassifierStatistic (int src, int dst)
int	Classify (const IntegralImage *intImage, const Rect &roi)
float	Evaluate (const IntegralImage *intImage, const Rect &roi, int indexWeak=-1)
WeakClassifier **	GetClassifierPool () const
void	SetClassifierPool (WeakClassifier **weaks)
int	GetNewBackup ()

Detailed Description

Select the best classifier from some weak classifiers. Here we only use WeakClassifierHaar. Used in online boosting classifier.

Author

Zhengrong Wang.

Constructor & Destructor Documentation

ClassifierSelector::ClassifierSelector	(	int	numW,
		const Size &	patchSize,
		int	numB = 2
	)

Build a classifier selector with some WeakClassifierHaars.

Parameters

numW	# weak classifiers
patchSize	the patch size used in weak classifiers
numB	# weak classifiers for subsititution

ClassifierSelector::ClassifierSelector	(	int	numW,
		WeakClassifier **	weaks,
		int	numB = 2
	)

Build a classifier selector from some outside weakclassifier array.

Parameters

weaks

the array of weak classifiers.

Member Function Documentation

int ClassifierSelector::Classify	(	const IntegralImage *	intImage,
		const Rect &	roi
	)

Evaluate a feature.

Returns

1 for pos, -1 for neg

float ClassifierSelector::GetError

(

int

index = -1

)

const

Returns

the error rate of this seletor, or any specific classifier.

void ClassifierSelector::Initialize

(

const Size &

patchSize

)

Initialize this selector for a new target. Notice the number of weak classifiers will not change, but all the features will be randomly generated again. This is designed to reduce the new/delete operation.

int ClassifierSelector::ReplaceWeakestClassifier ( float *  sumErrors, const Size &  patchSize  )

virtual

Replace the weakest classifier.

Parameters

sumErrors

a buffer contains the sum of error rate. The biggest weak classifier will be replaced

Returns

the index of the replaced classifier

void ClassifierSelector::ReplaceWeakestClassifierStatistic ( int  src, int  dst  )

virtual

Only replace the weight. Used when the weak classifier is from outside.

int ClassifierSelector::SelectBestClassifer ( float  importance, const bool *  errorMask, float *  errors  )

virtual

Select the best classifier.

Parameters

in	importance the weight of this sample
in	errorMask true if the classifer makes mistake on this sample
out	errors a buffer contains the error rates of each classifier

Returns

index of new selected classifier

void ClassifierSelector::Train	(	const IntegralImage *	intImage,
		const Rect &	roi,
		int	target,
		float	importance,
		bool *	errMask
	)

Train the weak classifiers.

Parameters

intImage	the integral image
roi	the region of the target
target	1 for pos, -1 for neg
importance	the weight of this sample
out	errMask update the error mask array

---

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

• include/ClassifierSelector.h
• src/ClassifierSelector.cpp