ossimKMeansClustering Class Reference

#include <ossimKMeansClustering.h>

Inheritance diagram for ossimKMeansClustering:

Classes
class	Cluster

Public Member Functions
	ossimKMeansClustering ()
	~ossimKMeansClustering ()
void	setNumClusters (ossim_uint32 K)
template<class T >
void	setSamples (T *samples, ossim_uint32 num_entries)
template<class T >
void	setPopulations (T *populations, ossim_uint32 num_entries)
bool	computeKmeans ()
ossim_uint32	getNumClusters () const
double	getMean (ossim_uint32 groupId) const
double	getSigma (ossim_uint32 groupId) const
double	getMinValue (ossim_uint32 groupId) const
double	getMaxValue (ossim_uint32 groupId) const
const ossimKMeansClustering::Cluster *	getCluster (ossim_uint32 i) const
void	setVerbose (bool v=true) const
Public Member Functions inherited from ossimReferenced
	ossimReferenced ()
	ossimReferenced (const ossimReferenced &)
ossimReferenced &	operator= (const ossimReferenced &)
void	ref () const
	increment the reference count by one, indicating that this object has another pointer which is referencing it. More...
void	unref () const
	decrement the reference count by one, indicating that a pointer to this object is referencing it. More...
void	unref_nodelete () const
	decrement the reference count by one, indicating that a pointer to this object is referencing it. More...
int	referenceCount () const

Private Attributes
ossim_uint32	m_numEntries
double *	m_samples
double *	m_populations
std::vector< Cluster >	m_clusters
bool	m_clustersValid
bool	m_verbose

Additional Inherited Members
Protected Member Functions inherited from ossimReferenced
virtual	~ossimReferenced ()

Detailed Description

Definition at line 25 of file ossimKMeansClustering.h.

Constructor & Destructor Documentation

ossimKMeansClustering()

ossimKMeansClustering::ossimKMeansClustering

(

)

Definition at line 13 of file ossimKMeansClustering.cpp.

:  m_numEntries(0),
   m_samples(0),
   m_populations(0),
   m_clustersValid(false),
   m_verbose(false)
{

}

~ossimKMeansClustering()

ossimKMeansClustering::~ossimKMeansClustering

(

)

Definition at line 23 of file ossimKMeansClustering.cpp.

References m_clusters, m_populations, and m_samples.

{
   delete m_samples;
   delete m_populations;
   m_clusters.clear();
}

Member Function Documentation

computeKmeans()

bool ossimKMeansClustering::computeKmeans

(

)

Definition at line 37 of file ossimKMeansClustering.cpp.

References m_clusters, m_clustersValid, m_numEntries, m_populations, m_samples, m_verbose, max, min, n, OSSIM_DEFAULT_MAX_PIX_DOUBLE, and OSSIM_DEFAULT_MIN_PIX_DOUBLE.

Referenced by ossimShorelineTool::autoComputeThreshold(), and ossimKMeansFilter::computeKMeans().

{
   if ((m_numEntries == 0) || (m_samples == 0))
         return false;

   if (m_verbose)
      std::cout.precision(6);

   // If populations aren't provided, assume only 1 of each sample:
   if (m_populations == 0)
   {
      m_populations = new double [m_numEntries];
      for (ossim_uint32 i=0; i<m_numEntries; ++i)
         m_populations[i] = 1.0;
   }

   // Scan for min and max:
   double overall_min = OSSIM_DEFAULT_MAX_PIX_DOUBLE;
   double overall_max = OSSIM_DEFAULT_MIN_PIX_DOUBLE;
   for (ossim_uint32 i=0; i<m_numEntries; i++)
   {
      if (m_populations[i] == 0)
         continue;

      if (m_samples[i] < overall_min)
         overall_min = m_samples[i];
      else if (m_samples[i] > overall_max)
         overall_max = m_samples[i];
   }

   double max_delta = overall_max - overall_min;
   // double convergenceThreshold = 0.1*(max_delta)/m_numEntries;
   ossim_uint32 numClusters = m_clusters.size();
   double* variances = new double [numClusters];
   ossim_uint32* priorCounts = new ossim_uint32 [numClusters];
   double interm_samples = (overall_max - overall_min) / numClusters;
   double mean_i = overall_min + interm_samples / 2.0;  // initial mean for cluster 0;
   double bound = overall_min;
   for (ossim_uint32 gid=0; gid<numClusters; ++gid)
   {
      //Initialize cluster with even spread and initial mean:
      m_clusters[gid].min = bound;
      bound += interm_samples;
      m_clusters[gid].max = bound;
      m_clusters[gid].mean = mean_i;
      m_clusters[gid].new_mean = 0;
      mean_i += interm_samples;
      m_clusters[gid].n = 0;
      m_clusters[gid].sigma = 1.0;
      variances[gid] = 0.0;
      priorCounts[gid] = 0;
   }
   double delta, min_delta;
   // ossim_uint32 best_gid, np, iters = 0, max_iters = 20;
   ossim_uint32 best_gid, iters = 0, max_iters = 20;
   bool converged = false;

   // Loop until converged on best solution:
   while (!converged && (iters < max_iters))
   {
      converged = true; // prove otherwise
      ++iters;

      for (ossim_uint32 gid=0; gid<numClusters; ++gid)
      {
         m_clusters[gid].min = overall_max;
         m_clusters[gid].max = overall_min;
      }

      for (ossim_uint32 i=0; i<m_numEntries; i++)
      {
         if ( m_populations[i] == 0)
            continue;

         // Find the current cluster:
         best_gid = 0;
         min_delta = OSSIM_DEFAULT_MAX_PIX_DOUBLE;
         for (ossim_uint32 gid=0; gid<numClusters; ++gid)
         {
            delta = fabs((m_samples[i] - m_clusters[gid].mean));
            if (delta < min_delta)
            {
               min_delta = delta;
               best_gid = gid;
            }
         }

         // Possible update of min/max for current cluster:
         if (m_samples[i] < m_clusters[best_gid].min)
            m_clusters[best_gid].min = m_samples[i];
         else if (m_samples[i] > m_clusters[best_gid].max)
            m_clusters[best_gid].max = m_samples[i];

         // Accumulate sample for the new mean for this cluster:
         m_clusters[best_gid].new_mean += m_populations[i]*m_samples[i];
         m_clusters[best_gid].n += m_populations[i];

         // Add this bin's contribution to the cluster stats. Use the sigma member as accumulator,
         // normalize later:
         delta = m_clusters[best_gid].mean - m_samples[i];
         variances[best_gid] += m_populations[i]*delta*delta;

      } // End loop over bins

      // Finished processing all input pixels for this iteration. Update the means:
      for (ossim_uint32 gid=0; gid<numClusters; ++gid)
      {
         // Compute new mean from accumulation:
         if (m_clusters[gid].n)
         {
            m_clusters[gid].new_mean /= m_clusters[gid].n;
            m_clusters[gid].sigma = sqrt(variances[gid] / m_clusters[gid].n);
            variances[gid] = 0.0;
         }

         //if (fabs(m_clusters[gid].mean - m_clusters[gid].new_mean) > convergenceThreshold )
         if (m_clusters[gid].n != priorCounts[gid])
            converged = false;

//         if (m_verbose)
//         {
//            cout<<"iteration: "<<iters<<endl;
//            cout<<"cluster["<<gid<<"].mean     = "<<m_clusters[gid].mean<<endl;
//            cout<<"cluster["<<gid<<"].new_mean = "<<m_clusters[gid].new_mean<<endl;
//            cout<<"cluster["<<gid<<"].sigma    = "<<m_clusters[gid].sigma<<endl;
//            cout<<"cluster["<<gid<<"].n        = "<<(int)m_clusters[gid].n<<"  prior: "<<priorCounts[gid]<<endl;
//            cout<<"cluster["<<gid<<"].min      = "<<m_clusters[gid].min<<endl;
//            cout<<"cluster["<<gid<<"].max      = "<<m_clusters[gid].max<<endl;
//            cout << endl;
//         }
         if (m_clusters[gid].n)
            m_clusters[gid].mean = m_clusters[gid].new_mean;

         if (!converged)
         {
            priorCounts[gid] = m_clusters[gid].n;
            m_clusters[gid].n = 0;
            m_clusters[gid].new_mean = 0;
         }
      }
   } // End overall loop for convergence:

   if (m_verbose)
   {
      cout<<"\nossimKMeansClustering Summary ("<<iters<<" iterations):"<<endl;
      for (ossim_uint32 gid=0; gid<numClusters; gid++)
      {
         cout<<"\n  cluster["<<gid<<"] n        = "<<(int)m_clusters[gid].n<<endl;
         cout<<"             mean     = "<<m_clusters[gid].mean<<endl;
         cout<<"             sigma    = "<<m_clusters[gid].sigma<<endl;
         cout<<"             min      = "<<m_clusters[gid].min<<endl;
         cout<<"             max      = "<<m_clusters[gid].max<<endl;
      }
      cout << endl;
   }

   delete [] variances;
   delete [] priorCounts;
   m_clustersValid = true;
   return true;

}

getCluster()

const ossimKMeansClustering::Cluster * ossimKMeansClustering::getCluster

(

ossim_uint32

i

)

const

Definition at line 233 of file ossimKMeansClustering.cpp.

References m_clusters.

Referenced by ossimKMeansFilter::getTile(), and ossimKMeansFilter::saveState().

{
   if (i >= m_clusters.size())
      return 0;
   return &(m_clusters[i]);
}

getMaxValue()

double ossimKMeansClustering::getMaxValue

(

ossim_uint32

groupId

)

const

Definition at line 224 of file ossimKMeansClustering.cpp.

References m_clusters, and ossim::nan().

{
   if (clusterId >= m_clusters.size())
      return ossim::nan();

   return m_clusters[clusterId].max;
}

getMean()

double ossimKMeansClustering::getMean

(

ossim_uint32

groupId

)

const

Definition at line 200 of file ossimKMeansClustering.cpp.

References m_clusters, and ossim::nan().

Referenced by ossimShorelineTool::autoComputeThreshold(), and ossimKMeansFilter::computeKMeans().

{
   if (clusterId >= m_clusters.size())
      return ossim::nan();

   return m_clusters[clusterId].mean;
}

getMinValue()

double ossimKMeansClustering::getMinValue

(

ossim_uint32

groupId

)

const

Definition at line 216 of file ossimKMeansClustering.cpp.

References m_clusters, and ossim::nan().

{
   if (clusterId >= m_clusters.size())
      return ossim::nan();

   return m_clusters[clusterId].min;
}

getNumClusters()

ossim_uint32 ossimKMeansClustering::getNumClusters ( ) const

inline

Definition at line 50 of file ossimKMeansClustering.h.

Referenced by ossimKMeansFilter::computeKMeans().

{ return m_clusters.size(); }

getSigma()

double ossimKMeansClustering::getSigma

(

ossim_uint32

groupId

)

const

Definition at line 208 of file ossimKMeansClustering.cpp.

References m_clusters, and ossim::nan().

Referenced by ossimShorelineTool::autoComputeThreshold(), and ossimKMeansFilter::computeKMeans().

{
   if (clusterId >= m_clusters.size())
      return ossim::nan();

   return m_clusters[clusterId].sigma;
}

setNumClusters()

void ossimKMeansClustering::setNumClusters

(

ossim_uint32

K

)

Definition at line 30 of file ossimKMeansClustering.cpp.

References m_clusters, and m_clustersValid.

Referenced by ossimShorelineTool::autoComputeThreshold(), and ossimKMeansFilter::computeKMeans().

{
   m_clusters.clear();
   m_clusters.resize(K);
   m_clustersValid = false;
}

setPopulations()

template<class T >

void ossimKMeansClustering::setPopulations	(	T *	populations,
		ossim_uint32	num_entries
	)

Definition at line 81 of file ossimKMeansClustering.h.

References m_clustersValid, and m_populations.

Referenced by ossimShorelineTool::autoComputeThreshold(), and ossimKMeansFilter::computeKMeans().

{
   if ((num_entries == 0) || (populations == 0))
      return;

   m_clustersValid = false;
   m_populations = new double[num_entries];
   for (ossim_uint32 i=0; i<num_entries; i++)
      m_populations[i] = (double) populations[i];
}

setSamples()

template<class T >

void ossimKMeansClustering::setSamples	(	T *	samples,
		ossim_uint32	num_entries
	)

Definition at line 69 of file ossimKMeansClustering.h.

References m_clustersValid, m_numEntries, and m_samples.

Referenced by ossimShorelineTool::autoComputeThreshold(), and ossimKMeansFilter::computeKMeans().

{
   if ((num_entries == 0) || (samples == 0))
      return;

   m_clustersValid = false;
   m_numEntries = num_entries;
   m_samples = new double[num_entries];
   for (ossim_uint32 i=0; i<num_entries; i++)
      m_samples[i] = (double) samples[i];
}

setVerbose()

void ossimKMeansClustering::setVerbose ( bool  v = true ) const

inline

Definition at line 58 of file ossimKMeansClustering.h.

Referenced by ossimKMeansFilter::computeKMeans().

{ m_verbose = v; }

Member Data Documentation

m_clusters

std::vector<Cluster> ossimKMeansClustering::m_clusters

private

Definition at line 64 of file ossimKMeansClustering.h.

Referenced by computeKmeans(), getCluster(), getMaxValue(), getMean(), getMinValue(), getSigma(), setNumClusters(), and ~ossimKMeansClustering().

m_clustersValid

bool ossimKMeansClustering::m_clustersValid

private

Definition at line 65 of file ossimKMeansClustering.h.

Referenced by computeKmeans(), setNumClusters(), setPopulations(), and setSamples().

m_numEntries

ossim_uint32 ossimKMeansClustering::m_numEntries

private

Definition at line 61 of file ossimKMeansClustering.h.

Referenced by computeKmeans(), and setSamples().

m_populations

double* ossimKMeansClustering::m_populations

private

Definition at line 63 of file ossimKMeansClustering.h.

Referenced by computeKmeans(), setPopulations(), and ~ossimKMeansClustering().

m_samples

double* ossimKMeansClustering::m_samples

private

Definition at line 62 of file ossimKMeansClustering.h.

Referenced by computeKmeans(), setSamples(), and ~ossimKMeansClustering().

m_verbose

bool ossimKMeansClustering::m_verbose

mutableprivate

Definition at line 66 of file ossimKMeansClustering.h.

Referenced by computeKmeans().

---

The documentation for this class was generated from the following files:

• ossimKMeansClustering.h
• ossimKMeansClustering.cpp