训练SVM分类器 VS2013 C++

#include "train_classifier.h" #include <fstream> #define DEBUG_OUT 1 #define ERROR 1 TrainClassifier::TrainClassifier() { //正样本数： _positive_sample_num = 0; //负样本数： _negative_sample_num = 0; //hog描述子参数： _nbins = 9; //hog梯度统计方向 _block_size = Size(16, 16); //滑动块大小 _block_stride = Size(8, 8); //滑动歩长 _cell_size = Size(8, 8); //hog统计块大小 } int TrainClassifier::getHogDescriptorLength(Size image_size) { return (size_t)_nbins * (_block_size.width / _cell_size.width) * (_block_size.height / _cell_size.height) * ((image_size.width - _block_size.width) / _block_stride.width + 1) * ((image_size.height - _block_size.height) / _block_stride.height + 1); } int TrainClassifier::generateClassifier(string save_classifier_path, string positive_sample_path,string negative_sample_path, Size image_size) { _hog_descriptor_length = getHogDescriptorLength(image_size); _image_size = image_size; //获取样本数： _positive_sample_num = getSampleNum(positive_sample_path); _negative_sample_num = getSampleNum(negative_sample_path); //样本总数： int total_sample_num = _positive_sample_num + _negative_sample_num; _sample_feature_mat = cvCreateMat(total_sample_num , _hog_descriptor_length, CV_32FC1); cvSetZero(_sample_feature_mat); //64*128的训练样本，该矩阵将是totalSample*3780,64*64的训练样本，该矩阵将是totalSample*1764 _sample_label_mat = cvCreateMat(total_sample_num, 1, CV_32FC1);//样本标识 cvSetZero(_sample_label_mat); //训练正样本： train(positive_sample_path, 1); //训练负样本: train(negative_sample_path, -1); //生成分类器： generator(save_classifier_path); return 0; } //获取样本数： int TrainClassifier::getSampleNum(string sample_path) { ifstream in_f; in_f.open(sample_path.c_str(), ios_base::in); if(in_f.is_open() == false) { #if DEBUG_OUT && ERROR cout<<"can't find file!"<<endl; #endif //DEBUG_OUT && ERROR return 1; } int sample_num = 0; string temp = ""; while(getline(in_f, temp)) { sample_num++; } in_f.close(); return sample_num; } //训练正负样本： int TrainClassifier::train(string sample_path, int label) { //从文件中读取数据： ifstream in_f; in_f.open(sample_path.c_str(), ios_base::in); if(in_f.is_open() == false) { #if DEBUG_OUT && ERROR cout<<"can't find file!"<<endl; #endif //DEBUG_OUT && ERROR return 1; } string image_path; static int i = 0; while(getline(in_f, image_path)) { cv::Mat image = imread(image_path); if(image.data != NULL) { cv::resize(image, image, _image_size, 0, 0, CV_INTER_LINEAR); cv::HOGDescriptor hog(_image_size, _block_size, _block_stride, _cell_size, _nbins); vector<float> featureVec; hog.compute(image, featureVec, _cell_size); int featureVecSize = featureVec.size(); for (int j=0; j<featureVecSize; j++) { CV_MAT_ELEM( *_sample_feature_mat, float, i, j ) = featureVec[j]; } _sample_label_mat->data.fl[i] = label; i++; } } in_f.close(); return 0; } //产生分类器： int TrainClassifier::generator(string save_classifier_path) { CvSVMParams params; params.svm_type = CvSVM::C_SVC; params.kernel_type = CvSVM::LINEAR; params.term_crit = cvTermCriteria(CV_TERMCRIT_ITER, 1000, FLT_EPSILON); params.C = 0.01; CvSVM svm; svm.train( _sample_feature_mat, _sample_label_mat, NULL, NULL, params ); //用SVM线性分类器训练 svm.save(save_classifier_path.c_str(), 0); cvReleaseMat(&_sample_feature_mat); cvReleaseMat(&_sample_label_mat); int supportVectorSize = svm.get_support_vector_count(); #if DEBUG_OUT cout<<"support num :"<<supportVectorSize<<endl; #endif //DEBUG_OUT return 0; }