at.mug.iqm.img.bundle.jai.stuff

Class KMeansImageClustering

java.lang.Object

java.lang.Thread

at.mug.iqm.img.bundle.jai.stuff.ImageProcessingTask

at.mug.iqm.img.bundle.jai.stuff.KMeansImageClustering

All Implemented Interfaces:

IProgressEmitter, ICancelable, IProcessingTask, Runnable

public class KMeansImageClustering
extends ImageProcessingTask

This class implements a basic K-Means clustering algorithm as an image processing task. This implementation tries to speed things up, but needs to keep all image data on memory. This implementation deals only with integer-like pixel data but can cluster N-dimensional data. One thing to remember about K-Means: it is possible that some clusters will become "empty". This implementation does not deal with any solution to this case, the clusters will simply become empty and will not be further processed. IMPORTANT: This implementation does not lock access to the data while the clustering is being done.

Changes

• 2009 08 Method: getClusterCenters added
• 2009 08 pixelArray[band] = (int)Math.round(clusterCenters[aCluster][band]); //Math.round added
• 2010 01 added progressbar support

Since:

2009 06 Original from Rafael Santos (rafael.santos@lac.inpe.br)

Author:

Helmut Ahammer

Nested Class Summary

Nested classes/interfaces inherited from class java.lang.Thread

Thread.State, Thread.UncaughtExceptionHandler

Field Summary

Fields

Modifier and Type	Field and Description
private int[]	aPixel
private int[]	clusterAssignmentCount
private float[][]	clusterCenters
private double	epsilon
private boolean	finished
private int	height
private int[]	inputData
private int	iteration
private int	maxIterations
private int	numBands
private int	numClusters
private short[][]	outputData
private long	position
private javax.media.jai.TiledImage	pOutput
private double	sumOfDistances
private int	width

Fields inherited from class at.mug.iqm.img.bundle.jai.stuff.ImageProcessingTask

isCancelable, parentTask, virtual

Fields inherited from class java.lang.Thread

MAX_PRIORITY, MIN_PRIORITY, NORM_PRIORITY

Constructor Summary

Constructors

Constructor and Description
KMeansImageClustering(javax.media.jai.PlanarImage pInput, int numClusters, int maxIterations, double epsilon, char initMode) The constructor for the class, which sets the input image, the number of desired clusters, the maximum number of iterations, a value that will be used to decide whether the convergence has stopped and the initial clusters centers initialization mode.

Method Summary

Modifier and Type	Method and Description
private short	getClassFor(int[] pixel) This auxiliary method gets the class for a pixel vector.
float[][]	getClusterCenters() This method returns the cluster centers.
private double[]	getExtrema(javax.media.jai.PlanarImage image) This method returns a two-element array with the minimum and maximum values found in a PlanarImage.
String	getInfo() This method returns some information about the task (its progress).
javax.media.jai.TiledImage	getOutput() This method returns the output image.
long	getPosition() This method returns a measure of the progress of the algorithm.
long	getSize() This method returns the estimated size (steps) for this task.
boolean	isFinished() This method returns true if the clustering has finished.
void	run() This method performs the bulk of the processing.

Methods inherited from class at.mug.iqm.img.bundle.jai.stuff.ImageProcessingTask

addProgressListener, fireProgressChanged, getParentTask, getPcs, getProgressListeners, isCancelable, isCancelled, isVirtual, setCancelable, setParentTask, setPcs, setVirtual

Methods inherited from class java.lang.Thread

activeCount, checkAccess, clone, countStackFrames, currentThread, destroy, dumpStack, enumerate, getAllStackTraces, getContextClassLoader, getDefaultUncaughtExceptionHandler, getId, getName, getPriority, getStackTrace, getState, getThreadGroup, getUncaughtExceptionHandler, holdsLock, interrupt, interrupted, isAlive, isDaemon, isInterrupted, join, join, join, resume, setContextClassLoader, setDaemon, setDefaultUncaughtExceptionHandler, setName, setPriority, setUncaughtExceptionHandler, sleep, sleep, start, stop, stop, suspend, toString, yield

Methods inherited from class java.lang.Object

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

Field Detail

pOutput

private javax.media.jai.TiledImage pOutput

width

private int width

height

private int height

numBands

private int numBands

maxIterations

private int maxIterations

numClusters

private int numClusters

iteration

private int iteration

sumOfDistances

private double sumOfDistances

epsilon

private double epsilon

finished

private boolean finished

position

private long position

clusterCenters

private float[][] clusterCenters

clusterAssignmentCount

private int[] clusterAssignmentCount

inputData

private int[] inputData

aPixel

private int[] aPixel

outputData

private short[][] outputData

Constructor Detail

KMeansImageClustering

public KMeansImageClustering(javax.media.jai.PlanarImage pInput,
                             int numClusters,
                             int maxIterations,
                             double epsilon,
                             char initMode)

The constructor for the class, which sets the input image, the number of desired clusters, the maximum number of iterations, a value that will be used to decide whether the convergence has stopped and the initial clusters centers initialization mode. It also allocates needed memory.

Parameters:

pInput - the input planar image.

numClusters - the desired number of clusters.

maxIterations - the maximum number of iterations.

epsilon - a small value used to verify if clustering has converged.

initMode - the initialization mode: 'R' for random values, 'S' for evenly spaced values, 'D' for randomly sampled data values.

Method Detail

getExtrema

private double[] getExtrema(javax.media.jai.PlanarImage image)

This method returns a two-element array with the minimum and maximum values found in a PlanarImage.

Parameters:

image - the input PlanarImage.

Returns:

an array with two positions: the first is the minima of the image and the second is the maxima.

run

public void run()

This method performs the bulk of the processing. It runs the classic K-Means clustering algorithm: 1 - Scan the image and calculate the assignment vector. 2 - Scan the assignment vector and recalculate the cluster centers. 3 - Calculate statistics and repeat from 1 if needed.

Specified by:

run in interface Runnable

Specified by:

run in class ImageProcessingTask

getClassFor

private short getClassFor(int[] pixel)

This auxiliary method gets the class for a pixel vector.

Returns:

the class (cluster index) for a pixel.

getOutput

public javax.media.jai.TiledImage getOutput()

This method returns the output image. It can return an image which is being processed while this method run, so results are not guaranteed to be final (and can be very strange if the image is large).

Returns:

the (possibly temporary) clustering results.

getSize

public long getSize()

This method returns the estimated size (steps) for this task. The value is, of course, an approximation, just so we will be able to give the user a feedback on the processing time. In this case, the value is calculated as the number of loops in the run() method.

Specified by:

getSize in class ImageProcessingTask

Returns:

the size of the image processing size.

getPosition

public long getPosition()

This method returns a measure of the progress of the algorithm.

Specified by:

getPosition in class ImageProcessingTask

Returns:

the position of the task processing.

isFinished

public boolean isFinished()

This method returns true if the clustering has finished.

Specified by:

isFinished in class ImageProcessingTask

Returns:

true if the processing task has finished.

getInfo

public String getInfo()

This method returns some information about the task (its progress).

Returns:

some information about the task

getClusterCenters

public float[][] getClusterCenters()

This method returns the cluster centers. added 2009 08

Returns:

cluster centers