C++ 实现k-means machine learning 算法 Computer Vision



C++实现k-means 算法， Machine Learning, Computer Vison

Hang_Kmeans.h 文件

 

#include "StdAfx.h"
#include "Hang_Kmeans.h"

Re_Kmeans::Re_Kmeans(void)
{
}

Re_Kmeans::~Re_Kmeans(void)
{
}

void Re_Kmeans::Inti_Cluser_Center(double*x_data,int x_num,int dim,int k_cluser,double*cer_point)
{
 //根据，要划分的聚类的群数，选择原始数据的索引位置点随机选取簇类中心

 set <int> A;
 set<int>::iterator iter;

 srand((int)time(0));

 //选取初始聚类中心

 for (int i=0;i<x_num;i++)
 { 
  int cluer_num=rand()%(x_num);
  if (A.size()<k_cluser)
  {
   A.insert(cluer_num);
  }
 } 

 vector<int> clu_number;

 for( iter=A.begin();iter!=A.end();iter++)
 {
  clu_number.push_back(*iter);
 }

 for (int i=0;i<clu_number.size();i++)
 {
  cer_point[i*dim+0]=x_data[clu_number[i]*dim+0];
  cer_point[i*dim+1]=x_data[clu_number[i]*dim+1];
 }

 A.clear();
 clu_number.clear();

}

void Re_Kmeans::Cluser_dis(double*x_data,int x_num,int dim,double*Inti_cerpoint,int k_cluser, vector<min_pos>&cluse_num)
{
 double*dis=new double[x_num*k_cluser];
 min_pos pos;

 for (int k=0;k<k_cluser;k++)  //计算每个点到初始类中心的距离
 {
  for (int i=0;i<x_num;i++)
  {
   dis[i*k_cluser+k]=sqrt(pow((x_data[i*dim+0]-Inti_cerpoint[k*dim+0]),2)+pow((x_data[i*dim+1]-Inti_cerpoint[k*dim+1]),2));
  }
 }

 for (int i=0;i<x_num;i++)   //分类
 {
  double dis_Int=dis[i*k_cluser+0];

  for (int k=0;k<k_cluser;k++)
  {
   if (dis[i*k_cluser+k]<=dis_Int)
   {
    dis_Int=dis[i*k_cluser+k];
    pos.dis=dis[i*k_cluser+k];
    pos.pointind=k;
   }
  }

  cluse_num.push_back(pos);
 }

 delete[] dis;

}

void Re_Kmeans::Re_cerpoint(double*x_data,int x_num,int dim,vector<min_pos>&cluse_num,int k_cluse,double*cer_pot)
{

 for (int k=0;k<k_cluse;k++)
 {
  int count=0;
  cer_pot[k*dim+0]=0.0;
  cer_pot[k*dim+1]=0.0;

  for (int i=0;i<x_num;i++)
  {
   if (cluse_num[i].pointind==k)
   {
    cer_pot[k*dim+0]=cer_pot[k*dim+0]+x_data[i*dim+0];
    cer_pot[k*dim+1]=cer_pot[k*dim+1]+x_data[i*dim+1];
    count=count+1;    
   }
  }
  cer_pot[k*dim+0]= cer_pot[k*dim+0]/count;
  cer_pot[k*dim+1]= cer_pot[k*dim+1]/count;
 }

}

void Re_Kmeans:: Chenk_point(double*cer_point,int k_cluser,int dim)
{
 Point pot;
 vector<Point> cerpot;

 double a=0.0;
 int Re_kluser=0;

 for (int i=0;i<k_cluser;i++)
 {
  if ((abs(cer_point[i*dim+0])<65535)&&(abs(cer_point[i*dim+1])<65535))
  {
   pot.x=cer_point[i*dim+0];
   pot.y=cer_point[i*dim+1];
   cerpot.push_back(pot);
   Re_kluser=Re_kluser+1;   
  }
  else
   continue;

 } 

 k_cluser=Re_kluser;

 for (int i=0;i<Re_kluser;i++)
 {
  cer_point[i*dim+0]=cerpot[i].x;
  cer_point[i*dim+1]=cerpot[i].y;
 }

 cerpot.clear();
}

void Re_Kmeans::Kmeans(double *x_data,int x_num,int dim,int k_cluser,double*cer_point)
{
 vector<min_pos>cluse_num;

 Inti_Cluser_Center(x_data,x_num,dim,k_cluser,cer_point);

 for(int iter=0;iter<20;++iter)
 {

  Cluser_dis(x_data,x_num,dim,cer_point,k_cluser,cluse_num);

  Re_cerpoint(x_data,x_num,dim,cluse_num,k_cluser,cer_point);

  cluse_num.erase(cluse_num.begin(),cluse_num.end());
 }

 //Chenk_point(cer_point,re_kluser,dim);

}

参考:  http://www.hangzh.com/