opencv3.0函数调用，svm训练分类器

#include <iostream>   

#include <fstream>   

#include <opencv2/opencv.hpp> 

#include <opencv2/ml/ml.hpp>

#include <opencv2/core/core.hpp>

#include <opencv2/highgui/highgui.hpp>

#include <opencv2/imgproc/imgproc.hpp>

#include <opencv2/objdetect/objdetect.hpp>

#include <string> 

#include <io.h>

using namespace std;

using namespace cv;

using namespace cv::ml;

#define PosSamNO 480    //正样本个数   

#define NegSamNO 3384  //负样本个数   

#define HardExampleNO 0  //难例个数 

int  main(int argc, char** argv)

{

       //HOG检测器，用来计算HOG描述子的 

       //检测窗口(48,48),块尺寸(16,16),块步长(8,8),cell尺寸(8,8),直方图bin个数9   param = CvSVMParams( CvSVM::C_SVC, CvSVM::RBF, 10.0, 0.09, 1.0, 10.0, 0.5, 1.0, NULL, criteria );

       cv::HOGDescriptor hog(cv::Size(64, 64), cv::Size(16, 16), cv::Size(16, 16), cv::Size(8, 8), 9);

       int DescriptorDim;//HOG描述子的维数，由图片大小、检测窗口大小、块大小、细胞单元中直方图bin个数决定  

       

       SVM::Params params;

       params.svmType = SVM::C_SVC;

       params.C = 0.1;

       params.kernelType = SVM::LINEAR;//SVM::LINEAR;//SVM::LINEAR;

       params.termCrit = TermCriteria(TermCriteria::MAX_ITER, (int)10000, 1e-6);

       //正样本图片路径文件，具体通过图片生成图片路径文件的代码，在我上传的资源里面可以找到

       std::ifstream finPos("E:/opencv_studydocumence/opencv_study/HOG+SVM/code/Release/labels_20180206/svm20180206_x.txt");

       //负样本图片的文件列表 

       std::ifstream finNeg("E:/opencv_studydocumence/opencv_study/HOG+SVM/code/Release/labels_20180206/svmtrain_neg.txt");

       std::string ImgName;

       //所有训练样本的特征向量组成的矩阵，行数等于所有样本的个数，列数等于HOG描述子维数  

       cv::Mat sampleFeatureMat;

       //训练样本的类别向量，行数等于所有样本的个数，列数等于1；1表示有目标，-1表示无目标  

       cv::Mat sampleLabelMat;

       //依次读取正样本图片，生成HOG描述子   

       for (int num = 0; num < PosSamNO && getline(finPos, ImgName); num++)

       {

              std::cout << "Processing：" << ImgName << std::endl;

              std::cout << "1" << std::endl;

              cv::Mat image = cv::imread(ImgName);

              //HOG描述子向量  

              std::vector<float> descriptors;

              //计算HOG描述子，检测窗口移动步长(8,8) 

              hog.compute(image, descriptors, cv::Size(8, 8));

              //处理第一个样本时初始化特征向量矩阵和类别矩阵，因为只有知道了特征向量的维数才能初始化特征向量矩阵   

              if (0 == num)

              {

                     //HOG描述子的维数  

                     DescriptorDim = descriptors.size();

                     //初始化所有训练样本的特征向量组成的矩阵，行数等于所有样本的个数，列数等于HOG描述子维数sampleFeatureMat   

                     sampleFeatureMat = cv::Mat::zeros(PosSamNO + NegSamNO + HardExampleNO, DescriptorDim, CV_32FC1);

                     //初始化训练样本的类别向量，行数等于所有样本的个数，列数等于1  

                     sampleLabelMat = cv::Mat::zeros(PosSamNO + NegSamNO + HardExampleNO, 1, CV_32SC1);

              }

              //将计算好的HOG描述子复制到样本特征矩阵sampleFeatureMat   

              for (int i = 0; i < DescriptorDim; i++)

              {

                     //第num个样本的特征向量中的第i个元素  

                     sampleFeatureMat.at<float>(num, i) = descriptors[i];

              }

              //正样本类别为17，有目标     

              sampleLabelMat.at<float>(num, 0) = 9;

       }

       //依次读取负样本图片，生成HOG描述子   

       for (int num = 0; num < NegSamNO && getline(finNeg, ImgName); num++)

       {

              std::cout << "-1"<< std::endl;

              std::cout << "Processing：" << ImgName << std::endl;

              cv::Mat src = cv::imread(ImgName);

              cv::resize(src, src, cv::Size(64, 64));

              //HOG描述子向量 

              std::vector<float> descriptors;

              //计算HOG描述子，检测窗口移动步长(8,8)  

              hog.compute(src, descriptors, cv::Size(8, 8));

              std::cout << "descriptor dimention：" << descriptors.size() << std::endl;

              //将计算好的HOG描述子复制到样本特征矩阵sampleFeatureMat   

              for (int i = 0; i < DescriptorDim; i++)

              {

                     //第PosSamNO+num个样本的特征向量中的第i个元素 

                     sampleFeatureMat.at<float>(num + PosSamNO, i) = descriptors[i];

              }

              //负样本类别为-1，无目标 

              sampleLabelMat.at<float>(num + PosSamNO, 0) = -1;

       }

       //处理HardExample负样本   

       if (HardExampleNO > 0)

       {

              //HardExample负样本的文件列表  

              std::ifstream finHardExample("svmtrain_hard.txt");

              //依次读取HardExample负样本图片，生成HOG描述子   

              for (int num = 0; num < HardExampleNO && getline(finHardExample, ImgName); num++)

              {

                     std::cout << "2" << std::endl;

                     std::cout << "Processing：" << ImgName << std::endl;

                     cv::Mat src = cv::imread(ImgName, cv::IMREAD_GRAYSCALE);

                     cv::resize(src, src, cv::Size(64, 64));

                     //HOG描述子向量   

                     std::vector<float> descriptors;

                     //计算HOG描述子，检测窗口移动步长(8,8)  

                     hog.compute(src, descriptors, cv::Size(8, 8));

                     //将计算好的HOG描述子复制到样本特征矩阵sampleFeatureMat   

                     for (int i = 0; i < DescriptorDim; i++)

                     {

                           //第PosSamNO+num个样本的特征向量中的第i个元素 

                           sampleFeatureMat.at<float>(num + PosSamNO + NegSamNO, i) = descriptors[i];

                     }

                     //负样本类别为-1，无目标  

                     sampleLabelMat.at<float>(num + PosSamNO + NegSamNO, 0) =2;

              }

       }

       //训练SVM分类器   

       //迭代终止条件，当迭代满1000次或误差小于FLT_EPSILON时停止迭代   

       std::cout << "start training SVM" << std::endl;

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

       Ptr<SVM> svm = StatModel::train<SVM>(sampleFeatureMat, ROW_SAMPLE, sampleLabelMat, params);

       cout << "Finished training process" << endl;

       //训练分类器   

       //svm->train(td);

       std::cout << "训练完成" << std::endl;

       //将训练好的SVM模型保存为xml文件 

       svm->save("20180206_xmls/svmx_20180206.xml");

       return 0;

}