vector_aggregation
Class AbstractFeatureAggregator

java.lang.Object
  vector_aggregation.AbstractFeatureAggregator

Direct Known Subclasses:

BowAggregator, VladAggregator, VladAggregatorMean, VladAggregatorWithFiltering

public abstract class AbstractFeatureAggregator
extends java.lang.Object

All methods which aggregate a set of local image descriptors should extend this abstract class.

Author:

Eleftherios Spyromitros-Xioufis

Field Summary

protected double[][]	codebook The codebook (centroids) used to aggregate the vectors.
protected int	descriptorLength The dimensionality of the local descriptors ( should be equal to the dimensionality of each centroid).
protected int	numCentroids The number of centroids in the codebook.
protected int	vectorLength The length of the generated vectors.

Constructor Summary

protected	AbstractFeatureAggregator()
protected	AbstractFeatureAggregator(double[][] codebook) The constructor.

Method Summary

double[]	aggregate(java.util.ArrayList<double[]> descriptors) This method performs some general checks before calling the aggregateInternal method which is implemented by each aggregator.
double[]	aggregate(double[][] descriptors) This method performs some general checks before calling the aggregateInternal method which is implemented by each aggregator.
protected abstract double[]	aggregateInternal(java.util.ArrayList<double[]> descriptors) This method should be overridden by all aggregators.
protected abstract double[]	aggregateInternal(double[][] descriptors) This method should be overridden by all aggregators.
protected int[]	computeKNearestCentroids(double[] descriptor, int k) Returns the indices of the k centroids which are closer to the given descriptor.
protected int	computeNearestCentroid(double[] descriptor) Returns the index of the centroid which is closer to the given descriptor.
protected double[]	computeNearestCentroidIndexAndDistance(double[] descriptor) Returns a double array which has the nearest centroid's index as the first element and the distance from this centroid as the second element.
int	getDescriptorLength()
int	getNumCentroids()
abstract int	getVectorLength() Should compute and return the length of the generated vector.
static double[][]	readCodebookFile(java.lang.String codebookFileName, int numCentroids, int centroidLength) Reads the codebook from the given file and returns it as two-dimensional double array.
void	setDescriptorLength(int descriptorLength)
void	setNumCentroids(int numCentroids)

Methods inherited from class java.lang.Object

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

Field Detail

codebook

protected double[][] codebook

The codebook (centroids) used to aggregate the vectors. Each centroid is stored in a different row.

numCentroids

protected int numCentroids

The number of centroids in the codebook.

vectorLength

protected int vectorLength

The length of the generated vectors.

descriptorLength

protected int descriptorLength

The dimensionality of the local descriptors ( should be equal to the dimensionality of each centroid).

Constructor Detail

AbstractFeatureAggregator

protected AbstractFeatureAggregator()

AbstractFeatureAggregator

protected AbstractFeatureAggregator(double[][] codebook)

The constructor.

Parameters:

codebook -

Method Detail

getVectorLength

public abstract int getVectorLength()

Should compute and return the length of the generated vector.

Returns:

getNumCentroids

public int getNumCentroids()

setNumCentroids

public void setNumCentroids(int numCentroids)

getDescriptorLength

public int getDescriptorLength()

setDescriptorLength

public void setDescriptorLength(int descriptorLength)

aggregate

public double[] aggregate(double[][] descriptors)
                   throws java.lang.Exception

This method performs some general checks before calling the aggregateInternal method which is implemented by each aggregator.

Parameters:

descriptors - a set of local descriptors

Returns:

a vector which aggregates the local descriptors

Throws:

java.lang.Exception

aggregateInternal

protected abstract double[] aggregateInternal(java.util.ArrayList<double[]> descriptors)
                                       throws java.lang.Exception

This method should be overridden by all aggregators.

Parameters:

descriptors -

Returns:

Throws:

java.lang.Exception

aggregate

public double[] aggregate(java.util.ArrayList<double[]> descriptors)
                   throws java.lang.Exception

This method performs some general checks before calling the aggregateInternal method which is implemented by each aggregator.

Parameters:

descriptors - a set of local descriptors

Returns:

a vector which aggregates the local descriptors

Throws:

java.lang.Exception

aggregateInternal

protected abstract double[] aggregateInternal(double[][] descriptors)
                                       throws java.lang.Exception

This method should be overridden by all aggregators.

Parameters:

descriptors -

Returns:

Throws:

java.lang.Exception

computeNearestCentroid

protected int computeNearestCentroid(double[] descriptor)

Returns the index of the centroid which is closer to the given descriptor.

Parameters:

descriptor -

Returns:

computeNearestCentroidIndexAndDistance

protected double[] computeNearestCentroidIndexAndDistance(double[] descriptor)

Returns a double array which has the nearest centroid's index as the first element and the distance from this centroid as the second element.

Parameters:

descriptor -

Returns:

computeKNearestCentroids

protected int[] computeKNearestCentroids(double[] descriptor,
                                         int k)

Returns the indices of the k centroids which are closer to the given descriptor. Can be used for soft quantization. Fast implementation with a bounded priority queue. TO DO: early stopping!

Parameters:

descriptor -

k -

Returns:

readCodebookFile

public static double[][] readCodebookFile(java.lang.String codebookFileName,
                                          int numCentroids,
                                          int centroidLength)
                                   throws java.io.IOException

Reads the codebook from the given file and returns it as two-dimensional double array.

Parameters:

codebookFileName - the name of the file containing the codebook

Throws:

java.io.IOException