HOG+SVM实现数字识别

#include <iostream> #include <opencv2/highgui.hpp> #include <opencv2/imgproc.hpp> #include "opencv2/objdetect.hpp" #include <opencv2/ml.hpp> using namespace cv::ml; using namespace cv; using namespace std; string pathName = "digits.png"; int SZ = 20; float affineFlags = WARP_INVERSE_MAP | INTER_LINEAR; //倾斜矫正 Mat deskew(Mat& img) { Moments m = moments(img); if (abs(m.mu02) < 1e-2) { return img.clone(); } float skew = m.mu11 / m.mu02; Mat warpMat = (Mat_<float>(2, 3) << 1, skew, -0.5*SZ*skew, 0, 1, 0); //2*3变换矩阵 Mat imgOut = Mat::zeros(img.rows, img.cols, img.type()); warpAffine(img, imgOut, warpMat, imgOut.size(), affineFlags); return imgOut; } //加载训练测试数据集标签 void loadTrainTestLabel(string &pathName, vector<Mat> &trainCells, vector<int> &trainLabels) { Mat img = imread(pathName, CV_LOAD_IMAGE_GRAYSCALE); int ImgCount = 0; for (int i = 0; i < img.rows; i = i + SZ) { for (int j = 0; j < img.cols; j = j + SZ) { Mat digitImg = (img.colRange(j, j + SZ).rowRange(i, i + SZ)).clone(); trainCells.push_back(digitImg); ImgCount++; } } cout << "Image Count : " << ImgCount << endl; float digitClassNumber = 0; for (int z = 0; z<ImgCount; z++) { if (z % 500 == 0 && z != 0) { digitClassNumber = digitClassNumber + 1; } trainLabels.push_back(digitClassNumber); } } // 创建倾斜矫正后的训练集测试集 void CreateDeskewedTrainTest(vector<Mat> &deskewedTrainCells, vector<Mat> &trainCells) { for (int i = 0; i<trainCells.size(); i++) { Mat deskewedImg = deskew(trainCells[i]); deskewedTrainCells.push_back(deskewedImg); } } //HOG特征描述子 HOGDescriptor hog( Size(20, 20), //winSize Size(10, 10), //blocksize Size(5, 5), //blockStride, Size(10, 10), //cellSize, 9, //nbins, 1, //derivAper, -1, //winSigma, 0, //histogramNormType, 0.2, //L2HysThresh, 0,//gammal correction, 64,//nlevels=64 1); //创建训练测试HOG特征，并将计算结果保存至 void CreateTrainTestHOG(vector<vector<float> > &trainHOG, vector<Mat> &deskewedtrainCells) { //train hog大小为num*hogsize for (int y = 0; y<deskewedtrainCells.size(); y++) { vector<float> descriptors; hog.compute(deskewedtrainCells[y], descriptors, Size(2, 2), Size(0, 0)); trainHOG.push_back(descriptors); } } void ConvertVectortoMatrix(vector<vector<float> > &trainHOG, Mat &trainMat) { int descriptor_size = trainHOG[0].size();//第一个样本的特征大小 for (int i = 0; i<trainHOG.size(); i++) { for (int j = 0; j<descriptor_size; j++) { trainMat.at<float>(i, j) = trainHOG[i][j]; } } } void getSVMParams(SVM *svm) { cout << "Kernel type : " << svm->getKernelType() << endl; cout << "Type : " << svm->getType() << endl; cout << "C : " << svm->getC() << endl; cout << "Degree : " << svm->getDegree() << endl; cout << "Nu : " << svm->getNu() << endl; cout << "Gamma : " << svm->getGamma() << endl; } void SVMtrain(Mat &trainMat, vector<int> &trainLabels) { Ptr<SVM> svm = SVM::create(); svm->setGamma(0.50625); svm->setC(12.5); svm->setKernel(SVM::RBF); svm->setType(SVM::C_SVC); Ptr<TrainData> td = TrainData::create(trainMat, ROW_SAMPLE, trainLabels); svm->train(td); //svm->trainAuto(td); svm->save("model4.yml"); getSVMParams(svm); } int main() { /*检测部分*/ /*Ptr<SVM> svm = Algorithm::load<SVM>("model4.yml");*/ //创建对象并初始化 Ptr<SVM> svm = SVM::create();//创建对象 svm = svm->load("model4.yml");//初始化 /*cout << svm->getVarCount() << endl;*/ if (!svm) { cout << "Load file failed..." << endl; } Mat img = imread(pathName, 0); if (img.empty()) { cout << "not exist" << endl; return -1; } Mat test = (img.colRange(0, 20).rowRange(60, 80)).clone(); /*imshow("te", test);*/ Mat skew = deskew(test); /*imshow("sk", skew);*/ vector<float> m_descriptor; hog.compute(skew, m_descriptor, Size(8, 8), Size(0, 0));//检测窗口移动步长 Mat img_hog(1, m_descriptor.size(), CV_32FC1); for (int i = 0; i < m_descriptor.size(); i++) { img_hog.at<float>(0, i) = m_descriptor[i]; } /*int r = svm->predict(m_descriptor);*/ int r =svm->predict(img_hog); cout << r << endl; /*训练部分*/ //vector<Mat> trainCells; // //vector<int> trainLabels; // //loadTrainTestLabel(pathName, trainCells,trainLabels); //vector<Mat> deskewedTrainCells; // //CreateDeskewedTrainTest(deskewedTrainCells, trainCells); //std::vector<std::vector<float> > trainHOG; // //CreateTrainTestHOG(trainHOG, deskewedTrainCells); //int descriptor_size = trainHOG[0].size(); //每张图片的特征数 //cout << "Descriptor Size : " << descriptor_size << endl; //Mat trainMat(trainHOG.size(), descriptor_size, CV_32FC1); // //ConvertVectortoMatrix(trainHOG, trainMat); //Mat testResponse; //SVMtrain(trainMat,trainLabels); waitKey(0); return 0; }