基于opencv的hog+svm简单分类实现

hog特征是一种进行物体检测的特征，常常被用来进行行人检测和物体分类。本文是利用opencv中的hog特征提取算法结合svm分类器进行简单的物体分类。

代码

#include <opencv2/core/core.hpp>
#include <opencv2/ml/ml.hpp>
#include <opencv2/features2d/features2d.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/nonfree/features2d.hpp>
#include <opencv2/legacy/legacy.hpp>
#include <opencv2/objdetect/objdetect.hpp>
#include <string>
#include <stdio.h>
#include <iostream>
using namespace cv;
using namespace std;
using namespace cv;
using namespace std;
int main()
{
    string classnum[5];
    classnum[0] = "./elephant/image_";
    classnum[1] = "./butterfly/image_";
    classnum[2] = "./cup/image_";
    classnum[3] = "./camera/image_";
    classnum[4] = "./crocodile_head/image_";
    vector<Mat> imageall;
    Mat data;
    const int trainnum = 45;
    const int testnum = 5;
    const int desnum = 30;
    const int k = 200;
    HOGDescriptor hog(Size(128, 128), Size(16, 16), Size(8, 8), Size(8, 8), 9);//定义hog描述子
    for (int i = 0; i < 5; i++)
    for (int j = 1; j <= trainnum; j++)
    {
        string num;
        char numm[10];
        sprintf(numm, "%04d.jpg", j);
        num = numm;
        string name = classnum[i] + num;
        cout << name << endl;
        Mat image1 = imread(name, 0);
        resize(image1, image1, Size(128, 128));
        cout <<"image:"<< image1.size() << endl;
        imshow("1", image1);
        vector<KeyPoint> keypoints1, keypoints2;
        //SurfFeatureDetector detector(desnum);
        //detector.detect(image1, keypoints1);
        //SiftDescriptorExtractor Desc;
        vector<float> descriptors;

        float fea[8100];
        //float fea[3459];
        hog.compute(image1, descriptors, Size(4, 4));//提取hog特征
        int dim = descriptors.size();
        cout << dim << endl;
        for (int i = 0; i < dim; i++)
            fea[i] = descriptors[i];
        data.push_back(Mat(1, dim, CV_32FC1, fea));


    }
    cout << data.size() << endl;
    Mat fealabel;
    for (int i = 0; i < trainnum*5; i++)
    {
        fealabel.push_back(Mat(1, 1, CV_32FC1, i / trainnum));
    }
    CvSVMParams params;
    params.svm_type = CvSVM::C_SVC;
    params.kernel_type = CvSVM::LINEAR;
    params.term_crit = cvTermCriteria(CV_TERMCRIT_ITER, 20000, 1e-6);

    // 对SVM进行训练
    CvSVM SVM;
    SVM.train(data, fealabel, Mat(), Mat(), params);
    for (int i = 0; i < 5 * trainnum; i++)
    {
        // cout<<feature.row(i)<<endl;
        cout << i / trainnum << " " << SVM.predict(data.row(i)) << endl;
    }
    Mat test;
    //测试过程
    for (int i = 0; i < 5; i++)
    for (int j =50-testnum+1; j <= 50; j++)
    {
        string num;
        char numm[10];
        sprintf(numm, "%04d.jpg", j);
        num = numm;
        string name = classnum[i] + num;
        cout << name << endl;
        Mat image1 = imread(name, 0);
        resize(image1, image1, Size(128, 128));
        cout << "image:" << image1.size() << endl;
        imshow("1", image1);
        vector<KeyPoint> keypoints1, keypoints2;
        SurfFeatureDetector detector(desnum);
        detector.detect(image1, keypoints1);
        SiftDescriptorExtractor Desc;
        vector<float> descriptors;

        float fea[8100];

        hog.compute(image1, descriptors, Size(4, 4));
        int dim = descriptors.size();
        cout << dim << endl;
        for (int i = 0; i < dim; i++)
            fea[i] = descriptors[i];
        test.push_back(Mat(1, dim, CV_32FC1, fea));


    }
    int count = 0;
    for (int i = 0; i < 5 * testnum; i++)
    {
        // cout<<feature.row(i)<<endl;
        int a = i / testnum;
        int b = SVM.predict(test.row(i));
        if (a == b)
            count++;
        cout << a << " " << b << endl;

    }
    cout << count*1.0 / (5 * testnum) << endl;
    waitKey();


}

注意hog特征在使用svm训练时一定要用linear的模型，我理解是用其他的模型容易产生过拟合。opencv svm参数是这样设置的：
params.kernel_type = CvSVM::LINEAR;
还有训练之前注意吧图片都resize成一样大小~