学习OpenCV——HOG+SVM

#include "cv.h"
#include "highgui.h"
#include "stdafx.h"
#include <ml.h>
#include <iostream>
#include <fstream>
#include <string>
#include <vector>
using namespace cv;
using namespace std;

int main(int argc, char** argv) 
{ 
    vector<string> img_path;
    vector<int> img_catg;
    int nLine = 0;
    string buf;
    ifstream svm_data( "E:/SVM_DATA.txt" );
    unsigned long n;

    while( svm_data )
    {
        if( getline( svm_data, buf ) )
        {
            nLine ++;
            if( nLine % 2 == 0 )
            {
                 img_catg.push_back( atoi( buf.c_str() ) );//atoi将字符串转换成整型，标志（0,1）
            }
            else
            {
                img_path.push_back( buf );//图像路径
            }
        }
    }
    svm_data.close();//关闭文件

    CvMat *data_mat, *res_mat;
    int nImgNum = nLine / 2;            //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, 1764, 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* src;
    IplImage* trainImg=cvCreateImage(cvSize(64,64),8,3);//需要分析的图片

    for( string::size_type i = 0; i != img_path.size(); i++ )
    {
            src=cvLoadImage(img_path[i].c_str(),1);
            if( src == NULL )
            {
                cout<<" can not load the image: "<<img_path[i].c_str()<<endl;
                continue;
            }

            cout<<" processing "<<img_path[i].c_str()<<endl;
            cvResize(src,trainImg);   //读取图片  
            HOGDescriptor *hog=new HOGDescriptor(cvSize(64,64),cvSize(16,16),cvSize(8,8),cvSize(8,8),9);  //具体意思见参考文章1,2  
            vector<float>descriptors;//结果数组  
            hog->compute(trainImg, descriptors,Size(1,1), Size(0,0)); //调用计算函数开始计算  
            cout<<"HOG dims: "<<descriptors.size()<<endl;
            //CvMat* SVMtrainMat=cvCreateMat(descriptors.size(),1,CV_32FC1);
            n=0;
            for(vector<float>::iterator iter=descriptors.begin();iter!=descriptors.end();iter++)
            {
                cvmSet(data_mat,i,n,*iter);
                n++;
            }
                //cout<<SVMtrainMat->rows<<endl;
            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 ); 
/*  
    SVM种类：CvSVM::C_SVC  
    Kernel的种类：CvSVM::RBF  
    degree：10.0（此次不使用）  
    gamma：8.0  
    coef0：1.0（此次不使用）  
    C：10.0  
    nu：0.5（此次不使用）  
    p：0.1（此次不使用）  
    然后对训练数据正规化处理，并放在CvMat型的数组里。  
                                                        */    
    //☆☆☆☆☆☆☆☆☆(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" ); 

    //检测样本
    IplImage *test;
    vector<string> img_tst_path;
    ifstream img_tst( "E:/SVM_TEST.txt" );
    while( img_tst )
    {
        if( getline( img_tst, buf ) )
        {
            img_tst_path.push_back( buf );
        }
    }
    img_tst.close();

    CvMat *test_hog = cvCreateMat( 1, 1764, CV_32FC1 );
    char line[512];
    ofstream predict_txt( "SVM_PREDICT.txt" );
    for( string::size_type j = 0; j != img_tst_path.size(); j++ )
    {
        test = cvLoadImage( img_tst_path[j].c_str(), 1);
        if( test == NULL )
        {
             cout<<" can not load the image: "<<img_tst_path[j].c_str()<<endl;
               continue;
         }
        cvZero(trainImg);
        cvResize(test,trainImg);   //读取图片  
        HOGDescriptor *hog=new HOGDescriptor(cvSize(64,64),cvSize(16,16),cvSize(8,8),cvSize(8,8),9);  //具体意思见参考文章1,2  
        vector<float>descriptors;//结果数组  
        hog->compute(trainImg, descriptors,Size(1,1), Size(0,0)); //调用计算函数开始计算  
        cout<<"HOG dims: "<<descriptors.size()<<endl;
        CvMat* SVMtrainMat=cvCreateMat(1,descriptors.size(),CV_32FC1);
        n=0;
        for(vector<float>::iterator iter=descriptors.begin();iter!=descriptors.end();iter++)
            {
                cvmSet(SVMtrainMat,0,n,*iter);
                n++;
            }

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

//cvReleaseImage( &src);
//cvReleaseImage( &sampleImg );
//cvReleaseImage( &tst );
//cvReleaseImage( &tst_tmp );
cvReleaseMat( &data_mat );
cvReleaseMat( &res_mat );

return 0;
}

【转】http://blog.csdn.net/sangni007/article/details/7471222