Class MeanShiftClustering

java.lang.Object

MeanShiftClustering

All Implemented Interfaces:

Clustering

public class MeanShiftClustering
extends java.lang.Object
implements Clustering

This class contains the main methods implementing the Mean-Shift clustering

Version:

jan 2014

Author:

Luca Castelli Aleardi and Steve Oudot (Ecole Polytechnique, INF562)

Field Summary

Fields

Modifier and Type	Field and Description
Cluster<Jcg.geometry.Point_d>	clusters
(package private) PointCloud	N
(package private) FastRangeSearch<Jcg.geometry.Point_d>	Rs
(package private) PointCloud	seeds
(package private) double	sqAvgRad
(package private) double	sqCvgRad
(package private) double	sqInflRad
(package private) double	sqMergeRad

Constructor Summary

Constructors

Constructor and Description
MeanShiftClustering(PointCloud n, double bandWidth)
MeanShiftClustering(PointCloud n, PointCloud s, double cr, double ar, double ir, double mr)

Method Summary

Methods

Modifier and Type	Method and Description
Jcg.geometry.Point_d[]	computeClusters() Main algorithm for detecting all clusters Clusters are detected iteratively (until all points are processed) Clusters are merged if required: when the corresponding peaks are close
PointCloud	detectCluster(Jcg.geometry.Point_d seed, int clusterIndex) Single cluster detection -- returns approximate peak and cluster points The output is a list of point containing: - all the points belonging to the detected cluster - the peak point (at the top of the list)
Cluster<Jcg.geometry.Point_d>	getClusters() Return the description of clusters
(package private) void	initMSC(PointCloud n, PointCloud s, double cr, double ar, double ir, double mr)
int	mergeCluster(PointCloud cluster, Jcg.geometry.Point_d[] clusterCenters) Perform cluster merging Remark: cluster center is at top of cluster cloud

Methods inherited from class java.lang.Object

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

Field Detail

clusters

public Cluster<Jcg.geometry.Point_d> clusters

N

PointCloud N

seeds

PointCloud seeds

sqCvgRad

double sqCvgRad

sqAvgRad

double sqAvgRad

sqInflRad

double sqInflRad

sqMergeRad

double sqMergeRad

Rs

FastRangeSearch<Jcg.geometry.Point_d> Rs

Constructor Detail

MeanShiftClustering

MeanShiftClustering(PointCloud n,
                   PointCloud s,
                   double cr,
                   double ar,
                   double ir,
                   double mr)

MeanShiftClustering

MeanShiftClustering(PointCloud n,
                   double bandWidth)

Method Detail

initMSC

void initMSC(PointCloud n,
           PointCloud s,
           double cr,
           double ar,
           double ir,
           double mr)

getClusters

public Cluster<Jcg.geometry.Point_d> getClusters()

Description copied from interface: Clustering

Return the description of clusters

Specified by:

getClusters in interface Clustering

Returns:

Cluster the description of the partition of points into clusters

detectCluster

public PointCloud detectCluster(Jcg.geometry.Point_d seed,
                       int clusterIndex)

Single cluster detection -- returns approximate peak and cluster points The output is a list of point containing: - all the points belonging to the detected cluster - the peak point (at the top of the list)

mergeCluster

public int mergeCluster(PointCloud cluster,
               Jcg.geometry.Point_d[] clusterCenters)

Perform cluster merging Remark: cluster center is at top of cluster cloud

Parameters:

cluster - the point cloud to be merge (cluster), containing at the top its center

clusterCenters[] - the set (array) of cluster centers

Returns:

-1 if no merge is performed, 'i' if the input cluster has been associated with the cluster i (already existing)

computeClusters

public Jcg.geometry.Point_d[] computeClusters()

Main algorithm for detecting all clusters Clusters are detected iteratively (until all points are processed) Clusters are merged if required: when the corresponding peaks are close

Specified by:

computeClusters in interface Clustering

Returns:

Point_d[] an array of points, of size n .First i elements are cluster centers (non null points), remaining n-i elements must be null