3. opencv进行SIFT特征提取

opencv中sift特征提取的步骤

1. 使用SiftFeatureDetector的detect方法检测特征存入一个向量里，并使用drawKeypoints在图中标识出来
2. SiftDescriptorExtractor 的compute方法提取特征描述符，特征描述符是一个矩阵
3. 使用匹配器matcher对描述符进行匹配，匹配结果保存由DMatch的组成的向量里
4. 设置距离阈值，使得匹配的向量距离小于最小距离的2被才能进入最终的结果，用DrawMatch可以显示

代码

// 使用Flann进行特征点匹配.cpp : 定义控制台应用程序的入口点。
//
#include "stdafx.h"
#include <opencv2/opencv.hpp>
#include <highgui/highgui.hpp>
#include <features2d/features2d.hpp>
#include <nonfree/nonfree.hpp>
#include <vector>
using namespace cv;
using namespace std;
int _tmain(int argc, _TCHAR* argv[])
{
    Mat input1 = imread("E://code//test//image//box.png", 1);
    Mat input2 = imread("E://code//test//image//box_in_scene.jpg", 1);
    if (input1.empty()||input2.empty())
    {
        cout << "不能正常加载图片" << endl;
        system("pause");
        return -1;
    }
    /************************************************************************/
    /*下面进行提取特征点*/
    /************************************************************************/
    SiftFeatureDetector feature;
    vector<KeyPoint> kerpoints1;
    feature.detect(input1, kerpoints1);
    Mat output1;
    drawKeypoints(input1, kerpoints1, output1);
    vector<KeyPoint> kerpoints2;
    feature.detect(input2, kerpoints2);
    Mat output2;
    drawKeypoints(input2, kerpoints2, output2);
    imshow("提取特征点后的box.png", output1);
    imshow("提取特征点后的box_in_scene.png", output2);
    imwrite("提取特征点后的box.png", output1);
    imwrite("提取特征点后的box_in_scene.png", output2);
    cout << "box提取的特征点数为:" << kerpoints1.size() << endl;
    cout << "box_in_scene的特征点数为:" << kerpoints2.size() << endl;
    /************************************************************************/
    /* 下面进行特征向量提取 */
    /************************************************************************/
    SiftDescriptorExtractor descript;
    Mat description1;
    descript.compute(input1, kerpoints1, description1);
    Mat description2;
    descript.compute(input2, kerpoints2, description2);
    /************************************************************************/
    /* 下面进行特征向量临近匹配 */
    /************************************************************************/
    vector<DMatch> matches;
    FlannBasedMatcher matcher;
    Mat image_match;
    matcher.match(description1, description2, matches);
    /************************************************************************/
    /* 下面计算向量距离的最大值与最小值 */
    /************************************************************************/
    double max_dist = 0, min_dist = 100;
    for (int i = 0; i < description1.rows; i++)
    {
        if (matches.at(i).distance>max_dist)
        {
            max_dist = matches[i].distance;
        }
        if (matches[i].distance<min_dist)
        {
            min_dist = matches[i].distance;
        }
    }
    cout << "最小距离为" << min_dist << endl;
    cout << "最大距离为" << max_dist << endl;
    /************************************************************************/
    /* 得到距离小于而V诶最小距离的匹配 */
    /************************************************************************/
    vector<DMatch> good_matches;
    for (int i = 0; i < matches.size(); i++)
    {
        if (matches[i].distance<2*min_dist)
        {
            good_matches.push_back(matches[i]);
            cout <<"第一个图中的"<< matches[i].queryIdx<<"匹配了第二个图中的"<<matches[i].trainIdx<<endl;
        }
    }
    drawMatches(input1, kerpoints1, input2, kerpoints2, good_matches, image_match);
    imshow("匹配后的图片", image_match);
    imwrite("匹配后的图片.png", image_match);
    cout << "匹配的特征点数为:" << good_matches.size() << endl;
    waitKey(0);
    return 0;
}

程序运行前的原始图片

提取特征点后

进行匹配后

相关代码介绍

    double max_dist = 0, min_dist = 100;
    for (int i = 0; i < description1.rows; i++)
    {
        if (matches.at(i).distance>max_dist)
        {
            max_dist = matches[i].distance;
        }
        if (matches[i].distance<min_dist)
        {
            min_dist = matches[i].distance;
        }
    }  

设置阈值，当特征向量的距离在最小距离的二倍范围内的，匹配为好的匹配;

本博文由时尚时尚最时尚的markdown编辑器编写.