（OpenCV3.2.0）基于HOG特征的SVM训练与测试实例

本人作为一个机器学习的小菜鸟，在学习相关知识的过程中，CSDN博客给了我很大的帮助，作为回馈，我也来写点博客为后来者铺路。

注意：在程序前需要包含一些必要的OpenCV头文件，我是将这些头文件写在了CvHeadFiles.h文件之中

#include  "CvHeadFiles.h"       

#include   <iostream>    

#include   <fstream>    

#include   <string>    

#include   <vector>  

using   namespace   cv;

using   namespace   cv::ml;

using   namespace   std;

int   main( int   argc ,   char **   argv )

{

         //--------------准备训练数据--------------------

         //定义相关变量

         int   ImgWidht = 24;

         int   ImgHeight = 24;

         vector < string > img_path;

         vector < int > img_catg;

         int   nLine = 0;

         string   buf;

         int   lab;

         ifstream   svm_data( "SVM_DATA.txt" );  

         unsigned   long   n;

         //依次读入样本的名和标签

         for (;;)

        {

               nLine++;

                if   (nLine % 2)

               {

                       svm_data   >>   buf;

                        if   (buf   ==   "end" )

                               break ;

                       img_path.push_back(buf); //图像路径  

               }

                else

               {

                       svm_data   >>   lab;

                       img_catg.push_back(lab);

               }

        }

        svm_data.close(); //关闭文件   

         //定义相关变量

         Mat   data_mat, res_mat;

         int   nImgNum = (nLine-1)/2;              //读入样本数量    

        res_mat   =   Mat ::zeros(nImgNum, 1,   CV_32SC1 );

         Mat   src;

         Mat   trainImg =   Mat ::zeros(ImgHeight, ImgWidht,   CV_8UC3 );    //需要分析的图片   

         //依次提取样本的HOG特征

         for   ( string :: size_type   i = 0; i != img_path.size(); i++)

        {

                //读入样本

               src   =   imread(img_path [ i ] .c_str(), 1);    //依次读入样本图片

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

               resize(src, trainImg, cv:: Size (ImgWidht, ImgHeight), 0, 0,   INTER_CUBIC );     //将样本图片调整为标准大小

               

                //对样本提取HOG特征向量

                HOGDescriptor   *hog =   new   HOGDescriptor (cvSize(ImgWidht, ImgHeight), cvSize(16, 16), cvSize(8, 8), cvSize(8, 8), 9);    //具体意思见参考文章1,2      

                vector < float > descriptors;    //存放提取HOG特征向量的结果数组 

               hog->compute(trainImg, descriptors,   Size (1, 1),   Size (0, 0));   //对读入的样本图片提取HOG特征向量 

                if   (i == 0)

               {

                       data_mat   =   Mat ::zeros(nImgNum, descriptors.size(),   CV_32FC1 );   //根据输入图片大小进行分配空间  

               }

               cout   <<   "HOG dims: "   <<   descriptors.size()   <<   endl;

                //存储所提取的样本的HOG特征值

               n = 0;

                for   ( vector < float >:: iterator   iter = descriptors.begin();iter   !=   descriptors.end();iter ++ )

               {

                       data_mat.at< float >(i, n) =   * iter;

                       n++;

               }

               

                //存储样本的标签

               res_mat.at< int >(i, 0) = img_catg [ i ] ;

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

        }

        

         //----------------------------设置SVM参数---------------------------------

        cout   <<   "\n------------------------------------------------------------\n" ;

        cout   <<   "Starting training process"   <<   endl;

         Ptr < SVM > svm =   SVM ::create();

        svm -> setType( SVM :: C_SVC );

        svm -> setC(10);

        svm -> setKernel( SVM :: RBF );

        svm -> setTermCriteria( TermCriteria ( TermCriteria :: MAX_ITER , 10000, 1e-6));

        

         //-------------------------------训练SVM------------------------------------------      

        svm -> train(data_mat,   ROW_SAMPLE , res_mat);     //注意此处data_mat和res_mat的类型

         //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   <<   "Finished training process"   <<   endl;

         //--------------------------------检测样本---------------------------------------   

         //依次读入测试样本名

         vector < string > img_tst_path;

         ifstream   img_tst( "SVM_TEST.txt" );

         for (;;)

        {

               img_tst   >>   buf;

                if   (buf   ==   "end" )

                        break ;

               img_tst_path.push_back(buf);

        }

        img_tst.close();

         //定义相关变量

         Mat   test;

         int   pos=0, neg=0,sum=0;

         char   line[512];

         ofstream   predict_txt( "SVM_PREDICT.txt" );

         for   ( string :: size_type   j = 0; j < img_tst_path.size(); j++)

        {

                //依次读入测试样本

               test   =   imread(img_tst_path [ j ] .c_str(), 1);     //读入图像    

               resize(test, trainImg, cv:: Size (ImgWidht, ImgHeight), 0, 0,   INTER_CUBIC );     //要搞成同样的大小才可以检测到        

               

                //提取测试样本的HOG特征

                HOGDescriptor   *hog =   new   HOGDescriptor (cvSize(ImgWidht, ImgHeight), cvSize(16, 16), cvSize(8, 8), cvSize(8, 8), 9);   

                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   <<   "The Detection Result:"   <<   endl;

               cout   <<   "HOG dims: "   <<   descriptors.size()   <<   endl;

                //存储所提取的特征

                Mat   SVMtrainMat =   Mat ::zeros(1, descriptors.size(),   CV_32FC1 );

               n = 0;

                for   ( vector < float >:: iterator   iter = descriptors.begin();iter   !=   descriptors.end();iter ++ )

               {

                       SVMtrainMat.at< float >(0, n) =   * iter;

                       n++;

               }

                //对提取的特征进行检验

                int   ret = svm -> predict(SVMtrainMat);

                if   (ret == 1)

                       pos++;

                else

                       neg++;

               std::sprintf(line,   "%s %d\r\n" , img_tst_path [ j ] .c_str(), ret);

               printf( "%s %d\r\n" , img_tst_path [ j ] .c_str(), ret);

                //getchar();

               predict_txt   <<   line;

        }

        sum = neg + pos;

        cout   <<   "neg="   <<   neg   <<   "\n" ;

        cout   <<   "pos="   <<   pos   <<   "\n" ;

        cout   <<   "sum="   <<   sum   <<   "\n" ;

        predict_txt.close();

        system( "pause" );

         return   0;

}