svm图像训练与测试

#include <cv.h>
#include <highgui.h>
#include <ml.h>
#include<boost/tokenizer.hpp>
#include<boost/lexical_cast.hpp>

#include <iostream>
#include <fstream>

#include <string>
#include <vector>
using namespace std;
using namespace boost;
#define WIDTH 20
#define HEIGHT 20

int main( )
{
    vector<string> img_path;
    vector<int> img_catg;

    string buf;
    ifstream svm_data( "/home/dz/data/label/train" );
    string spath="/home/dz/data/imageS/";
    char_separator<char> sep(";");
    while( svm_data )
    {
        if( getline( svm_data, buf ) )
        {
          tokenizer<char_separator<char> > tokens(buf,sep);
          tokenizer<char_separator<char> >::iterator it;
          it=tokens.begin();
          img_path.push_back( spath+*it+".jpg" );++it;
            img_catg.push_back( lexical_cast<int>(*it) );//atoi将字符串转换成整型，标志（0,1）
        }
    }
    svm_data.close();//关闭文件


    CvMat *data_mat, *res_mat;
    int nImgNum = img_path.size();            //读入样本数量
    样本矩阵，nImgNum：横坐标是样本数量， WIDTH * HEIGHT：样本特征向量，即图像大小
    data_mat = cvCreateMat( nImgNum, WIDTH * HEIGHT, CV_32FC1 );
    cvSetZero( data_mat );
    //类型矩阵,存储每个样本的类型标志
    res_mat = cvCreateMat( nImgNum, 1, CV_32FC1 );
    cvSetZero( res_mat );

    IplImage *srcImg, *sampleImg;
    float b;
    unsigned  n;

    for( string::size_type i = 0; i != img_path.size(); i++ )
    {
       srcImg = cvLoadImage( img_path[i].c_str(), CV_LOAD_IMAGE_GRAYSCALE );
       if( srcImg == NULL )
       {
            cout<<" can not load the image: "<<img_path[i].c_str()<<endl;
            continue;
       }

       cout<<" processing "<<img_path[i].c_str()<<endl;

       sampleImg = cvCreateImage( cvSize( WIDTH, HEIGHT ), IPL_DEPTH_8U, 1 );//样本大小（WIDTH, HEIGHT）
       cvResize( srcImg, sampleImg );//改变图像大小

       cvSmooth( sampleImg, sampleImg );    //降噪
       //生成训练数据
       n = 0;
        for( int ii = 0; ii < sampleImg->height; ii++ )
        {
            for( int jj = 0; jj < sampleImg->width; jj++, n++ )
            {
                // b = (float)((int)((uchar)( sampleImg->imageData + sampleImg->widthStep * ii + jj )) / 255.0 );
                b=((float*)(sampleImg->imageData+sampleImg->widthStep*ii))[jj];
                 cvmSet( data_mat, (int)i, n, b );
            }
        }
        cvmSet( res_mat, i, 0, img_catg[i] );
        cout<<" end processing "<<img_path[i].c_str()<<" "<<img_catg[i]<<endl;
    }


    CvSVM svm = CvSVM();
    CvSVMParams param;
    CvTermCriteria criteria;
    criteria = cvTermCriteria( CV_TERMCRIT_EPS, 1000, FLT_EPSILON );
    param = CvSVMParams( CvSVM::C_SVC, CvSVM::RBF, 10.0, 0.09, 1.0, 10.0, 0.5, 1.0, NULL, criteria );
    
    //☆☆☆☆☆☆☆☆☆(5)SVM学习☆☆☆☆☆☆☆☆☆☆☆☆
    svm.train( data_mat, res_mat, NULL, NULL, param );
    //☆☆利用训练数据和确定的学习参数,进行SVM学习☆☆☆☆
    svm.save( "SVM_DATA.xml" );
cout<<"adsf "<<endl;

    //检测样本
    IplImage *tst, *tst_tmp;
    vector<string> img_tst_path;
    ifstream img_tst( "/home/dz/data/label/test" );
    while( img_tst )
    {
        if( getline( img_tst, buf ) )
        {
                tokenizer<char_separator<char> > tokens(buf,sep);
          tokenizer<char_separator<char> >::iterator it;
          it=tokens.begin();
          img_tst_path.push_back( spath+*it+".jpg" );

        }
    }
    img_tst.close();

    CvMat *tst_mat = cvCreateMat( 1, WIDTH*HEIGHT, CV_32FC1 );
    char line[512];
    ofstream predict_txt( "SVM_PREDICT.txt" );
    for( string::size_type j = 0; j != img_tst_path.size(); j++ )
    {
        tst = cvLoadImage( img_tst_path[j].c_str(), CV_LOAD_IMAGE_GRAYSCALE );
        if( tst == NULL )
        {
             cout<<" can not load the image: "<<img_tst_path[j].c_str()<<endl;
               continue;
   }
   tst_tmp = cvCreateImage( cvSize( WIDTH, HEIGHT ), IPL_DEPTH_8U, 1 );
   cvResize( tst, tst_tmp );
   cvSmooth( tst_tmp, tst_tmp );
   n = 0;
   for(int ii = 0; ii < tst_tmp->height; ii++ )
   {
     for(int jj = 0; jj < tst_tmp->width; jj++, n++ )
     {
         //b = (float)(((int)((uchar)tst_tmp->imageData+tst_tmp->widthStep*ii+jj))/255.0);
         b=((float*)(tst_tmp->imageData+tst_tmp->widthStep*ii))[jj];
         cvmSet( tst_mat, 0, n, (double)b );
     }
   }

   int ret = svm.predict( tst_mat );
   sprintf( line, "%s %d\r\n", img_tst_path[j].c_str(), ret );
   predict_txt<<line;
}
predict_txt.close();

cvReleaseImage( &srcImg );
cvReleaseImage( &sampleImg );
cvReleaseImage( &tst );
cvReleaseImage( &tst_tmp );
cvReleaseMat( &data_mat );
cvReleaseMat( &res_mat );

return 0;

}