学习OpenCV——一次颜色聚类

1.首先区分三个通道，画出每个通道的直方图；

2.平滑直方图，像素值等于前后共七种颜色数量的平均值；

3.取波峰，波谷。（条件：颜色数亮s(i-2)<s(i-1)<s(i)>s(i+1)>s(i+2)或s(i-2)>s(i-1)>s(i)<s(i+1)<s(i+2)））

4.计算算有颜色的组合，并存入一个一维矩阵中；

5.颜色合并（利用map<int,int>）；

6.把原图所有颜色与合并后的颜色计算距离，得出距离最近的合并颜色，在另一个原图大小的1通道矩阵中，记录相应位置颜色对应合并颜色的序号；

7.把原图序号相同的颜色加和，求平均值，就是类聚一次后的颜色。

8.把聚类后的颜色存入相应的序号的位置。

即为一次聚类后的结果，效果不很好。是由于聚类次数过少。

利用Kmean2算法，效果会很好。

 

#include "cv.h"#include"highgui.h"#include<map>#include<vector>#include<iostream>using namespace std;template<typename T>T sqr(T x){return x*x;}CvMat* mat=0;IplImage* image=0;IplImage *histimg=cvCreateImage(cvSize(256,300),IPL_DEPTH_8U,3);IplImage* DrawHist(map<int,int>colornum,int n);map<int,int> Fine(CvMat* mat1u);//CvMat* Merge(CvMat* Fine,int FineNum);int main(){image=cvLoadImage("D:/7.jpg",1);cvNamedWindow("Show",1);cvShowImage("Show",image);IplImage *hist0=0;//cvCreateImage(cvSize(300,300),IPL_DEPTH_8U,1);IplImage *hist1=0;//cvCreateImage(cvSize(300,300),IPL_DEPTH_8U,1);CvMat* r=cvCreateMat(image->height,image->width,CV_8UC1);CvMat* g=cvCreateMat(image->height,image->width,CV_8UC1);CvMat* b=cvCreateMat(image->height,image->width,CV_8UC1);CvMat* rgb=cvCreateMat(image->height,image->width,CV_32FC3);CvMat* result=cvCreateMat(image->height,image->width,CV_32FC3);CvMat* result00=cvCreateMat(image->height,image->width,CV_8UC3);cvConvertScale(image,rgb,1.0/255);cvSplit(image,r,g,b,NULL);map<int,int>fine_r=Fine(r);map<int,int>fine_g=Fine(g);map<int,int>fine_b=Fine(b);int FineNum=fine_r.size()*fine_g.size()*fine_b.size();cout<<"FineNum: "<<FineNum<<endl;CvMat* Fine=cvCreateMat(1,FineNum,CV_32FC3);//vector<CvScalar>avg;////////////////////////////int i=0;for(map<int,int>::iterator it0=fine_r.begin();it0!=fine_r.end();it0++){for(map<int,int>::iterator it1=fine_g.begin();it1!=fine_g.end();it1++){float* f=(float*)(Fine->data.ptr);for(map<int,int>::iterator it2=fine_b.begin();it2!=fine_b.end();it2++){f[3*i]=(float)it0->first/255;f[3*i+1]=(float)it1->first/255;f[3*i+2]=(float)it2->first/255;i++;}}}//CvMat* Fine2=0;//Fine2=Merge(Fine,FineNum);vector<pair<int,int>>merge;float* f1=(float*)(Fine->data.ptr);float* f2=(float*)(Fine->data.ptr);for(int m=0;m<FineNum;m++){for(int n=m+1;n<FineNum;n++){float d=sqr(f1[3*m]-f2[3*n])+sqr(f1[3*m+1]-f2[3*n+1])+sqr(f1[3*m+2]-f2[3*n+2]);if(d<=0.1){merge.push_back(make_pair(m,n));}}}cout<<"merge:"<<merge.size()<<endl;//Fine1中存储把颜色距离平方小于0.05的元素合并后的结果，结果存在第一个元素位置，第二个位置置零CvMat* Fine1=cvCreateMat(1,FineNum,CV_32FC3);float* F1=(float*)(Fine1->data.ptr);for(vector<pair<int,int>>::iterator iter=merge.begin();iter!=merge.end();iter++){if(F1[3*(iter->first)]!=0){F1[3*(iter->first)]=(f1[3*(iter->first)]+f1[3*(iter->second)])/2;F1[3*(iter->first)+1]=(f1[3*(iter->first)+1]+f1[3*(iter->second)+1])/2;F1[3*(iter->first)+2]=(f1[3*(iter->first)+2]+f1[3*(iter->second)+2])/2;F1[3*(iter->second)]=0;F1[3*(iter->second)+1]=0;F1[3*(iter->second)+2]=0;f1[3*(iter->second)]=0;f1[3*(iter->second)+1]=0;f1[3*(iter->second)+2]=0;}else{F1[3*(iter->first)]=f1[3*(iter->first)];F1[3*(iter->first)+1]=f1[3*(iter->first)+1];F1[3*(iter->first)+2]=f1[3*(iter->first)+2];}}//求Fine1中非零元素个数int no=0;for(int m=0;m<FineNum;m++){if(F1[m]!=0)no++;}cout<<"Not Zore Num:"<<no<<endl;//Fine2中存储merge后Fine1中的非零元素CvMat* Fine2=cvCreateMat(1,no*3,CV_32FC1);float* F2=(float*)(Fine2->data.ptr);for(int m=0;m<FineNum;m++){if(F1[m]!=0){*F2=F1[m];F2++;}}cout<<"F2_Width: "<<Fine2->width<<endl;//int no=Fine2->width;CvMat* id =cvCreateMat(image->height,image->width,CV_32SC1);for(int r=0;r<rgb->rows;r++){int* idx=(int*)(id->data.ptr+r*id->step);float* rgbData=(float*)(rgb->data.ptr+r*rgb->step);for(int c=0;c<rgb->cols;c++,idx++){float D=200000;int t=0;float* f=(float*)(Fine2->data.ptr);for(int i=0;i<no;i++){float dist=sqr(rgbData[3*c]-f[3*i])+sqr(rgbData[3*c+1]-f[3*i+1])+sqr(rgbData[3*c+2]-f[3*i+2]);if(dist<D){D=dist;t=i;}*idx=t;}}}//vector<CvScalar>avg1;//////////////////////CvMat *colorNum= cvCreateMat(1,no, CV_32SC1);cvZero(Fine2);cvZero(colorNum);float* f=(float*)(Fine2->data.ptr);int* num1=(int*)(colorNum->data.ptr);for(int r=0;r<rgb->rows;r++){int* idx=(int*)(id->data.ptr+r*id->step);float* rgbData=(float*)(rgb->data.ptr+r*rgb->step); for(int c=0;c<rgb->cols;c++){f[3*idx[c]]+=rgbData[3*c];f[3*idx[c]+1]+=rgbData[3*c+1];f[3*idx[c]+2]+=rgbData[3*c+2];num1[*idx]++;}}for(int i=0;i<no;i++){f[3*i]/=num1[i];f[3*i+1]/=num1[i];f[3*i+2]/=num1[i];}for(int r=0;r<rgb->rows;r++){int* idx=(int*)(id->data.ptr+r*id->step);float* resultData=(float*)(result->data.ptr+r*result->step);for(int c=0;c<rgb->cols;c++,idx++){//CvScalar avg1;resultData[3*c]=f[3*(*idx)];resultData[3*c+1]=f[3*(*idx)+1];resultData[3*c+2]=f[3*(*idx)+2];//cvSet2D(result,r,c,avg1);}}//cvConvertScale(result,result00,255);cvNamedWindow("Result",1);cvShowImage("Result",result);//cvSaveImage("D:/test/out_C-mean量化颜色(峰&谷，直接合并)+(d小于0.1)merge颜色/9.jpg",result00);cvWaitKey(0);}map<int,int> Fine(CvMat* mat1u){map<int,int>colornum1;IplImage *hist0=0;//cvCreateImage(cvSize(300,300),IPL_DEPTH_8U,1);IplImage *hist1=0;//cvCreateImage(cvSize(300,300),IPL_DEPTH_8U,1);for(int i=0;i<mat1u->rows;i++){for(int j=0;j<mat1u->cols;j++){int color=(int)(cvGetReal2D(mat1u,i,j));++colornum1[color];}}int n=colornum1.size();/*colornum1[0]=0;colornum1[1]=0;colornum1[2]=0;colornum1[n-3]=0;colornum1[n-1]=0;colornum1[n-2]=0;*/cout<<"ColorNum:"<<n<<endl;hist0=DrawHist(colornum1,n);cvNamedWindow("Hist0",1);cvShowImage("Hist0",hist0);//平滑vector<pair<int, int>> num1;for( map<int,int>::iterator iter=colornum1.begin();iter!=colornum1.end();iter++){num1.push_back(pair<int,int>(iter->first,iter->second));}for(vector<pair<int,int>>::size_type i=3;i!=n-4;i++){int n=(int)((num1[i-3].second+num1[i-2].second+num1[i-1].second+num1[i].second+num1[i+1].second+num1[i+2].second+num1[i+3].second)/7);colornum1[num1[i].first]=n;}hist1=DrawHist(colornum1,n);num1.clear();for( map<int,int>::iterator iter=colornum1.begin();iter!=colornum1.end();iter++){num1.push_back(pair<int,int>(iter->first,iter->second));}map<int,int>merge1;map<int,int>merge2;int j=0;for(vector<pair<int,int>>::size_type i=3;i!=n-4;i++){if((num1[i-2].second<num1[i-1].second&&num1[i-1].second<num1[i].second && num1[i].second>num1[i+1].second&& num1[i+1].second>num1[i+2].second||num1[i-2].second>num1[i-1].second&&num1[i-1].second>num1[i].second && num1[i].second<num1[i+1].second&& num1[i+1].second<num1[i+2].second)&&num1[i-1].second!=0&&num1[i-2].second!=0&&num1[i-3].second!=0&&num1[i+1].second!=0&&num1[i+2].second!=0&&num1[i+3].second!=0){merge1[i]=num1[i].second;cout<<num1[i].first<<": "<<num1[i].second<<endl;}}//////////////////////////////////////////////////////////////////////////////////////同一通道上颜色距离小于32的删除vector<pair<int, int>> num3;for( map<int,int>::iterator iter=merge1.begin();iter!=merge1.end();iter++){num3.push_back(pair<int,int>(iter->first,iter->second));}vector<pair<int,int>>::iterator it3,it2;it2=num3.begin();it3=num3.begin();++it3;while(it3!=num3.end()){if(abs(it2->first-it3->first)<32){it3=num3.erase(it3);}else{it2=it3;it3++;}}int m3=num3.size();cout<<"merge color num:"<<m3<<endl;for(vector<pair<int,int>>::size_type i=0;i!=m3;i++){cout<<num3[i].first<<endl;merge2[num3[i].first]=num3[i].second;}/////////////////////////////////////////////////////////////////////////////////////////////for( map<int,int>::iterator iter=merge2.begin();iter!=merge2.end();iter++){int x=iter->first;double y=iter->second;int h=(int)(300-0.1*y);cvCircle(hist1,cvPoint(x,h),1,CV_RGB(255,0,0),-1);}cvNamedWindow("Hist1",1);cvShowImage("Hist1",hist1);cvWaitKey(0);return merge2;}IplImage* DrawHist(map<int,int>colornum,int n){cvZero(histimg);int bin_w=1;double bin_h=(double)300/3000;for(map<int,int>::iterator it=colornum.begin();it!=colornum.end();it++){ double val =(double) (it->second*bin_h );        CvScalar color = CV_RGB(255,255,0); //(hsv2rgb(i*180.f/hdims);        cvLine( histimg, cvPoint(it->first*bin_w,300),cvPoint((it->first)*bin_w,(int)(300-val)),color, 1, 8, 0 );}    return histimg;}