1 enum Types {
 2 /** C-Support Vector Classification. n-class classification (n \f$\geq\f$ 2), allows
 3 imperfect separation of classes with penalty multiplier C for outliers. */
 4 C_SVC=100, //C类支持向量分类机。 n类分组  (n≥2)，允许用异常值惩罚因子C进行不完全分类。
 5 /** \f$\nu\f$-Support Vector Classification. n-class classification with possible
 6 imperfect separation. Parameter \f$\nu\f$ (in the range 0..1, the larger the value, the smoother
 7 the decision boundary) is used instead of C. */
 8 NU_SVC=101,// 类支持向量分类机。n类似然不完全分类的分类器。参数为取代C（其值在区间【0，1】中，nu越大，决策边界越平滑）。
 9 /** Distribution Estimation (One-class %SVM). All the training data are from
10 the same class, %SVM builds a boundary that separates the class from the rest of the feature
11 space. */
12 ONE_CLASS=102,//单分类器，所有的训练数据提取自同一个类里，然后SVM建立了一个分界线以分割该类在特征空间中所占区域和其它类在特征空间中所占区域。
13 /** \f$\epsilon\f$-Support Vector Regression. The distance between feature vectors
14 from the training set and the fitting hyper-plane must be less than p. For outliers the
15 penalty multiplier C is used. */
16 EPS_SVR=103,//类支持向量回归机。 代替了 p。
17 /** \f$\nu\f$-Support Vector Regression. \f$\nu\f$ is used instead of p.
18 See @cite LibSVM for details. */
19 NU_SVR=104
20 };

 1 enum KernelTypes {
 2         /** Returned by SVM::getKernelType in case when custom kernel has been set */
 3         CUSTOM=-1,
 4         /** Linear kernel. No mapping is done, linear discrimination (or regression) is
 5         done in the original feature space. It is the fastest option. \f$K(x_i, x_j) = x_i^T x_j\f$. */
 6         LINEAR=0,//线性核函数
 7         /** Polynomial kernel:
 8         \f$K(x_i, x_j) = (\gamma x_i^T x_j + coef0)^{degree}, \gamma > 0\f$. */
 9         POLY=1,//多项式函数(r*u'v + coef0)^degree
10         /** Radial basis function (RBF), a good choice in most cases.
11         \f$K(x_i, x_j) = e^{-\gamma ||x_i - x_j||^2}, \gamma > 0\f$. */
12         RBF=2,//exp(-gamma|u-v|^2)基于径向的函数，对于大多数情况都是一个较好的选择：
13         /** Sigmoid kernel: \f$K(x_i, x_j) = \tanh(\gamma x_i^T x_j + coef0)\f$. */
14         SIGMOID=3,//tanh(r*u'v + coef0)
15         /** Exponential Chi2 kernel, similar to the RBF kernel:
16         \f$K(x_i, x_j) = e^{-\gamma \chi^2(x_i,x_j)}, \chi^2(x_i,x_j) = (x_i-x_j)^2/(x_i+x_j), \gamma > 0\f$. */
17         CHI2=4,
18         /** Histogram intersection kernel. A fast kernel. \f$K(x_i, x_j) = min(x_i,x_j)\f$. */
19         INTER=5
20     };

1  TermCriteria::TermCriteria(int _type, int _maxCount, double _epsilon)
2     : type(_type), maxCount(_maxCount), epsilon(_epsilon) {}
3     //_type:迭代结束的类型，如果为CV_TERMCRIT_ITER，则用最大迭代次数作为终止条件，如果为CV_TERMCRIT_EPS 则用精度作为迭代条件，如果为CV_TERMCRIT_ITER+CV_TERMCRIT_EPS则用最大迭代次数或者精度作为迭代条件，看哪个条件先满足。
4     //_maxCount：终止程序的最大迭代次数
5     //_epsilon：终止程序的精度

1     train( InputArray samples, int layout, InputArray responses );
2     //samples：输入数据
3     //layout： 读取数据方式
4                     //----ROW_SAMPLE = 0, //!< 一行为一个样本
5                     //----COL_SAMPLE = 1  //!< 一列为一个样本
6     //responses：返回的数据（默认为不返回）

  1 #include <opencv2/opencv.hpp>
  2 #include <iostream>
  3 #include "math.h"
  4
  5 using namespace cv;
  6 using namespace std;
  7 int test = 0;
  8 RNG rng(99999);
  9 void drawCross(Mat &img, Point center, Scalar color)
 10 {
 11     int col = center.x > 2 ? center.x : 2;
 12     int row = center.y> 2 ? center.y : 2;
 13     line(img, Point(col - 2, row - 2), Point(col + 2, row + 2), color);
 14     line(img, Point(col + 2, row - 2), Point(col - 2, row + 2), color);
 15 }
 16
 17 void newSvmTest(int rows, int cols, int testCount)
 18 {
 19     assert(testCount >= rows * cols);
 20
 21     Mat img = Mat::zeros(rows, cols, CV_8UC3);
 22     Mat testPoint = Mat::zeros(rows, cols, CV_8UC1);
 23     Mat data = Mat::zeros(testCount, 2, CV_32FC1);
 24     Mat res = Mat::zeros(testCount, 1, CV_32SC1);
 25
 26     //Create random test points
 27     for (int i = 0; i < testCount; i++)
 28     {
 29         int row = rng.uniform(0,255) % rows;
 30         int col = rng.uniform(0,255) % cols;
 31         //-----存储数据
 32         if (testPoint.at<unsigned char>(row, col) == 0)
 33         {
 34             testPoint.at<unsigned char>(row, col) = 1;
 35             data.at<float>(i, 0) = float(col) / cols;
 36             data.at<float>(i, 1) = float(row) / rows;
 37         }
 38         else//防止重复
 39         {
 40             i--;
 41             continue;
 42         }
 43         //分成三类
 44         if (row > (50 * cos(col * CV_PI / 100) + 200))
 45         {
 46             drawCross(img, Point(col, row), CV_RGB(255, 0, 0));
 47             res.at<unsigned int>(i, 0) = 1;
 48         }
 49         else
 50         {
 51             if (col > 200)
 52             {
 53                 drawCross(img, Point(col, row), CV_RGB(0, 255, 0));
 54                 res.at<unsigned int>(i, 0) = 2;
 55             }
 56             else
 57             {
 58                 drawCross(img, Point(col, row), CV_RGB(0, 0, 255));
 59                 res.at<unsigned int>(i, 0) = 3;
 60             }
 61         }
 62
 63     }
 64     Mat show = img.clone();
 65     //----SVM训练
 66     cv::Ptr<cv::ml::SVM> svm = cv::ml::SVM::create();
 67     svm->setType(cv::ml::SVM::Types::C_SVC);
 68     svm->setKernel(cv::ml::SVM::KernelTypes::RBF);
 69     svm->setDegree(3);
 70     svm->setGamma(3);
 71     svm->setCoef0(1.0);
 72     svm->setC(10);
 73     svm->setNu(0);
 74     svm->setP(0.1);
 75     svm->setTermCriteria(cv::TermCriteria(cv::TermCriteria::EPS, 1000, FLT_EPSILON));
 76     // train operation
 77     svm->train(data, cv::ml::SampleTypes::ROW_SAMPLE, res);
 78     //----预测数据
 79     for (int i = 0; i< rows; i++)
 80     {
 81         for (int j = 0; j< cols; j++)
 82         {
 83             Mat m = Mat::zeros(1, 2, CV_32FC1);
 84             m.at<float>(0, 0) = float(j) / cols;
 85             m.at<float>(0, 1) = float(i) / rows;
 86
 87             float ret = 0.0;
 88             ret = svm->predict(m);
 89             Scalar rcolor;
 90
 91             switch ((int)ret)
 92             {
 93             case 1: rcolor = Scalar(100, 0, 0); break;
 94             case 2: rcolor = Scalar(0, 100, 0); break;
 95             case 3: rcolor = Scalar(0, 0, 100); break;
 96             }
 97
 98             line(img, Point(j, i), Point(j, i), rcolor);
 99         }
100     }
101     //-------获得支持向量
102     int thickness = 2;
103     int lineType = 8;
104     //Mat sv = svm->getUncompressedSupportVectors();//不能使用，不知道什么原因，网上有的人可以使用
105     Mat sv = svm->getSupportVectors();
106     for (int i = 0; i < sv.rows; ++i)
107     {
108         const float* v = sv.ptr<float>(i);
109         circle(img, Point((int)(v[0]*600), (int)(v[1]*400)), 4, Scalar(255, 255, 255), thickness, lineType);
110     }
111
112     imshow("dst", img);
113     waitKey(0);
114 }
115
116 int main(int argc,char** argv[])
117 {
118     newSvmTest(400, 600, 500);
119     return 0;
120 }