opencv中的Linear SVM例子

SVM(Support Vector Machine)作为经典的分类算法，运用广泛，可以通过正的样本和负的样本训练，从而达到预测新的样本属于正的还是负的。线性SVM实际上找到一个超平面，该超平面到正负样本的距离最大，从而实现最优的二分类。这里介绍opencv自带的例子introduction_to_svm.cpp，本人对其稍作修改，完整的代码如下：

#include <opencv2/core.hpp>
#include <opencv2/imgproc.hpp>
#include "opencv2/imgcodecs.hpp"
#include <opencv2/highgui.hpp>
#include <opencv2/ml.hpp>


#include <opencv2/opencv.hpp>


using namespace cv;
using namespace cv::ml;
using namespace std;


int main(int, char**)
{
    // Data for visual representation
    int width = 512, height = 512;
    Mat image = Mat::zeros(height, width, CV_8UC3);


    // Set up training data
    //! [setup1]
    int labels[4] = {1, 1, -1, -1};
    float trainingData[4][2] = { {501, 10}, {255, 80}, {501, 255}, {10, 501} };
    //! [setup1]
    //! [setup2]
    Mat trainingDataMat(4, 2, CV_32FC1, trainingData);
    Mat labelsMat(4, 1, CV_32SC1, labels);
    //! [setup2]

    // Train the SVM
    //! [init]
    Ptr<SVM> svm = SVM::create();
    svm->setType(SVM::C_SVC);
    svm->setKernel(SVM::LINEAR);
    svm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 100, 1e-6));
    //! [init]
    //! [train]
    svm->train(trainingDataMat, ROW_SAMPLE, labelsMat);
    //! [train]


    // Show the decision regions given by the SVM
    //! [show]
    Vec3b green(0,255,0), blue (255,0,0);
    for (int i = 0; i < image.rows; ++i)
        for (int j = 0; j < image.cols; ++j)
        {
            Mat sampleMat = (Mat_<float>(1,2) << j,i);
            float response = svm->predict(sampleMat);


            if (response == 1)
                image.at<Vec3b>(i,j)  = green;
            else if (response == -1)
                image.at<Vec3b>(i,j)  = blue;
        }
    //! [show]


    // Show the training data
    //! [show_data]
    int thickness = -1;
    int lineType = 8;
for (int i = 0; i < sizeof(trainingData)/sizeof(trainingData[0]); i++)
{
circle( image, Point(trainingData[i][0], trainingData[i][1]), 5, labels[i] == 1 ? Scalar(255, 255, 255) : Scalar(0, 0, 0), thickness, lineType );
}
    //! [show_data]


    // Show support vectors
    //! [show_vectors]
    thickness = 2;
    lineType  = 8;
    Mat sv = svm->getSupportVectors();
    for (int i = 0; i < sv.rows; ++i)
    {
        const float* v = sv.ptr<float>(i);
cout << "[" << v[0] << ", " << v[1] << "]'" << endl;
    }
    //! [show_vectors]

    vector<int> compression_params;
    compression_params.push_back(IMWRITE_PNG_COMPRESSION);
    compression_params.push_back(9);


    try {
        imwrite("result.png", image, compression_params);        // save the image
    }
    catch (runtime_error& ex) {
        fprintf(stderr, "Exception converting image to PNG format: %s\n", ex.what());
        return 1;
    }

    imshow("SVM Simple Example", image); // show it to the user
    waitKey(0);
}

运行后，得到support vector: [0.00232288, 0.00816326]'

和下图：