ILClassifier Interface Reference

This interface provide functionalities of the machine learning based image classification. More...

Inheritance diagram for ILClassifier:

Public Member Functions
void	AddClass (LString classLabel, ILSampleDatabase *sampleDB)
int	AddFeature (ILFeature *feature)
int	AddModel (ILModel *model)
int	CountFeature ()
int	CountModel ()
void	CreateClass (LString classLabel, LString sampleDBDir)
ILSampleDatabase *	GetClassSampleDb (LString classLabel)
ILFeature *	GetFeature (int idx)
ILModel *	GetModel (int idx)
BOOL	IsTrained ()
LPVErrorCode	Predict (ILImage *img, ILRegion *predictRegion, LString *classLabel, double *confidence)
void	RemoveAllClass (BOOL deleteSampleDb)
void	RemoveAllFeatures ()
void	RemoveAllModels ()
void	RemoveClass (LString classLabel, BOOL deleteSampleDb)
void	RemoveFeature (int idx)
void	RemoveModel (int idx)
LPVErrorCode	Train ()
Public Member Functions inherited from ILObject
ILObject *	Copy ()
LPVErrorCode	Load (LString filename)
void	Reset ()
LPVErrorCode	Save (LString filename)
BOOL	Valid ()

Properties
int	PrimaryModelIndex [get, set]
	The index of the primary model which is used for prediction. By default, we use the first model.
LString	WorkingDir [get, set]
	The working directory, used as the default root path for the database of the created class labels.

Detailed Description

This interface provide functionalities of the machine learning based image classification.

This is an integrated image classifier which may involve multiple machine learning models with various feature extraction approaches. You may train those models with one Train() call, then choose the primary one for the prediction, according to their testing performance.

We also provide the advanced functionalities for auto-tuning, auto-feature-selection and powerful testing.

To use this interface, you should create a LClassifier object.

Example Code

C++

ILClassifierPtr llcPtr = LClassifier::Create();
llcPtr->WorkingDir = L"d:/test";
llcPtr->AddClass(L"label1", db1);
llcPtr->AddClass(L"label2", db2);
llcPtr->AddFeature(hog);
llcPtr->AddModel(svm);
// train
LPVErrorCode err = llcPtr->Train();
if (err != LPVErrorCode::LPVNoError || !llcPtr->IsTrained()) return;
// predict
std::wstring resultLabel;
double resultConf = 0;
err = llcPtr->Predict(testImg, testRegion, &resultLabel, &resultConf);
if (err != LPVErrorCode::LPVNoError) return;

C#

ILClassifierPtr llcPtr = new LClassifier();
llcPtr.WorkingDir = "d:/test";
llcPtr.AddClass("label1", db1);
llcPtr.AddClass("label2", db2);
llcPtr.AddFeature(hog);
llcPtr.AddModel(svm);
// train
LPVErrorCode err = llcPtr.Train();
if (err != LPVErrorCode.LPVNoError || !llcPtr.IsTrained()) return;
// predict
string resultLabel;
double resultConf = 0;
err = llcPtr.Predict(testImg, testRegion, out resultLabel, out resultConf);
if (err != LPVErrorCode.LPVNoError) return;

COM

ILClassifierPtr llcPtr = LClassifier::Create();
// or, ILClassifierPtr llcPtr(__uuidof(LClassifier)), for absence of built-in Create functions (before 2.2.0)
llcPtr->WorkingDir = L"d:/test";
llcPtr->AddClass(L"label1", db1);
llcPtr->AddClass(L"label2", db2);
llcPtr->AddFeature(hog);
llcPtr->AddModel(svm);
// train
LPVErrorCode err = llcPtr->Train();
if (err != LPVErrorCode::LPVNoError || !llcPtr->IsTrained()) return;
// predict
CComBSTR resultLabel;
double resultConf = 0;
err = llcPtr->Predict(testImg, testRegion, &resultLabel, &resultConf);
if (err != LPVErrorCode::LPVNoError) return;

Member Function Documentation

AddClass()

void AddClass	(	LString	classLabel,
		ILSampleDatabase *	sampleDB
	)

Add a new class and the input associated sample database.

Parameters

[in]	classLabel	The label of the class. It may be a readable word and must be not duplicated.
[in]	sampleDB	The associated sample database.

See also

CreateClass(), RemoveClass()

AddFeature()

int AddFeature

(

ILFeature *

feature

)

Add the feature to the classifier.

Parameters

[in]

feature

The input feature object.

Return values

idx	Return the index of the new added feature.

See also

GetFeature(), RemoveFeature()

AddModel()

int AddModel

(

ILModel *

model

)

Add the model to the classifier.

Parameters

[in]

model

The input model object.

Return values

idx	Return the index of the new added model.

See also

GetModel(), RemoveModel()

CountFeature()

int CountFeature

(

)

Get the total count of all the features.

Return values

val	The total count.

CountModel()

int CountModel

(

)

Get the total count of all the models.

Return values

val	The total count.

CreateClass()

void CreateClass	(	LString	classLabel,
		LString	sampleDBDir
	)

Create a new class and its sample database.

Parameters

[in]	classLabel	The label of the class. It may be a readable word and must be not duplicated.
[in]	sampleDBDir	The required directory for the class' sample database. It could be empty then we'll create one under the default path generated out of WorkingDir, or a path to an existing database.

See also

AddClass(), RemoveClass()

GetClassSampleDb()

ILSampleDatabase* GetClassSampleDb

(

LString

classLabel

)

Get the sample database of the required class.

Parameters

[in]

classLabel

The label of the class.

Return values

sampleDB

Return the sample database. It may be null if there's no matching label.

GetFeature()

ILFeature* GetFeature

(

int

idx

)

Get the feature object with the given index.

Parameters

[in]

idx

The input index, it should be a 0-based number smaller than total count.

Return values

feature

Return the feature object.

See also

AddFeature(), CountFeature()

GetModel()

ILModel* GetModel

(

int

idx

)

Get the model object with the given index.

Parameters

[in]

idx

The input index, it should be a 0-based number smaller than total count.

Return values

model

Return the model object.

See also

AddModel(), CountModel()

IsTrained()

BOOL IsTrained

(

)

Check whether the classifier is well-trained

Return values

val	Return True if it's trained, otherwise, return False.

Predict()

LPVErrorCode Predict	(	ILImage *	img,
		ILRegion *	predictRegion,
		LString *	classLabel,
		double *	confidence
	)

Predict the input image with the region, report its class label.

Parameters

[in]	img	The input image
[in]	predictRegion	The input prediction region, could be null.
[out]	classLabel	Output the result class label.
[out]	confidence	Output the confidence of this prediction. It's a numeric positive value mostly in 0 ~ 1, the higher indicates the classifier is more confident and the result is more trustworthy.

Return values

error

Return error code if anything is wrong.

RemoveAllClass()

void RemoveAllClass

(

BOOL

deleteSampleDb

)

Remove all the classes.

Parameters

[in]

deleteSampleDb

Whether to permanently delete all the sample databases together. By default, we don't.

RemoveAllFeatures()

void RemoveAllFeatures

(

)

Remove all the features in the classifier

See also

RemoveFeature()

RemoveAllModels()

void RemoveAllModels

(

)

Remove all the models in the classifier

See also

RemoveModel()

RemoveClass()

void RemoveClass	(	LString	classLabel,
		BOOL	deleteSampleDb
	)

Remove the class from the classifier.

Parameters

[in]	classLabel	The label of the class.
[in]	deleteSampleDb	Whether to permanently delete the sample database together. By default, we don't.

RemoveFeature()

void RemoveFeature

(

int

idx

)

Remove the feature with the given index.

Parameters

[in]

idx

The input index, it should be a 0-based number smaller than total count.

See also

RemoveAllFeatures(), AddFeature()

RemoveModel()

void RemoveModel

(

int

idx

)

Remove the model with the given index.

Parameters

[in]

idx

The input index, it should be a 0-based number smaller than total count.

See also

RemoveAllModels(), AddModel()

Train()

LPVErrorCode Train

(

)

Train the classifier. All the features in it will engage in, and all the models will be trained.

Return values

error

Return error code if anything is wrong.