diva.sketch.classification

Class WeightedEuclideanClassifier

java.lang.Object

diva.sketch.classification.AbstractClassifier

diva.sketch.classification.WeightedEuclideanClassifier

All Implemented Interfaces:

Classifier, TrainableClassifier

public class WeightedEuclideanClassifier
extends AbstractClassifier

WeightedEuclideanClassifier is a trainable classifier that uses a weighted N-dimensional distance between feature sets to classify its input.

Version:

$Revision: 1.7 $

Author:

Heloise Hse (hwawen@eecs.berkeley.edu), Michael Shilman (michaels@eecs.berkeley.edu)

Field Summary

Fields

Modifier and Type	Field and Description
protected static double	MIN_SIGMA The minimum sigma value; used to avoid divide-by-zero errors.
protected static double	NORMALIZATION A normalization constant: 10 divided by 30, every 30 unit in distance results in a 10% recognition drop.

Fields inherited from class diva.sketch.classification.AbstractClassifier

_weights

Constructor Summary

Constructors

Constructor and Description
WeightedEuclideanClassifier() Construct a classifier with no weight set.

Method Summary

Modifier and Type	Method and Description
Classification	classify(FeatureSet fs) Classify the specified feature set using each weight, by comparing them to the mu (mean) value of the weight and weighting it by the sigma value (standard deviation).

Methods inherited from class diva.sketch.classification.AbstractClassifier

clear, debug, isIncremental, train

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

MIN_SIGMA

protected static final double MIN_SIGMA

The minimum sigma value; used to avoid divide-by-zero errors.

See Also:

Constant Field Values

NORMALIZATION

protected static final double NORMALIZATION

A normalization constant: 10 divided by 30, every 30 unit in distance results in a 10% recognition drop.

See Also:

Constant Field Values

Constructor Detail

WeightedEuclideanClassifier

public WeightedEuclideanClassifier()

Construct a classifier with no weight set. The weight set is created by the train method.

Method Detail

classify

public Classification classify(FeatureSet fs)
                        throws ClassifierException

Classify the specified feature set using each weight, by comparing them to the mu (mean) value of the weight and weighting it by the sigma value (standard deviation). For each feature f,

    value = sum((input[f] - mu[f])^2/sigma[f]^2)
 

Finally, normalize the value into a confidence between 0 and 100.

Throws:

ClassifierException