一.特征提取与匹配（ORB）

#include <iostream>

#include <opencv2/core/core.hpp>

#include <opencv2/features2d/features2d.hpp>

#include <opencv2/highgui/highgui.hpp>

using namespace std;

using namespace cv;

int main ( int argc, char** argv )

{

    if ( argc != 3 )

    {

        cout<<"usage: feature_extraction img1 img2"<<endl;

        return 1;

    }      //运行方式

///

//-- 读取图像

    Mat img_1 = imread ( argv[1], CV_LOAD_IMAGE_COLOR );

    Mat img_2 = imread ( argv[2], CV_LOAD_IMAGE_COLOR );

 //-- 初始化特征点与描述子

    std::vector<KeyPoint> keypoints_1, keypoints_2;

class KeyPoint

{

Point2f  pt;  //特征点坐标

float  size; //特征点邻域直径

float  angle; //特征点的方向，值为0~360，负值表示不使用

float  response; //特征点的响应强度，代表了该点是特征点的程度，可以用于后续处理中特征点排序

int  octave; //特征点所在的图像金字塔的组

int  class_id; //用于聚类的id

}

    Mat descriptors_1, descriptors_2;

    Ptr<FeatureDetector> detector = ORB::create();

    Ptr<DescriptorExtractor> descriptor = ORB::create();

    // Ptr<FeatureDetector> detector = FeatureDetector::create(detector_name);

    // Ptr<DescriptorExtractor> descriptor = DescriptorExtractor::create(descriptor_name);

    //暴力匹配法

    Ptr<DescriptorMatcher> matcher  = DescriptorMatcher::create ( "BruteForce-Hamming" );

 //-- 第一步:检测 Oriented FAST 角点位置

    detector->detect ( img_1,keypoints_1 );

    detector->detect ( img_2,keypoints_2 );

 //-- 第二步:根据角点位置计算 BRIEF 描述子

    descriptor->compute ( img_1, keypoints_1, descriptors_1 );

    descriptor->compute ( img_2, keypoints_2, descriptors_2 );

//--绘制特征点

    Mat outimg1;

//画有特征点的图像drawKeypoints函数

    drawKeypoints( img_1, keypoints_1, outimg1, Scalar::all(-1), DrawMatchesFlags::DEFAULT );

    imshow("ORB特征点",outimg1);

 //-- 第三步:对两幅图像中的BRIEF描述子进行匹配，使用 Hamming 距离

    vector<DMatch> matches;

    //BFMatcher matcher ( NORM_HAMMING );

    matcher->match ( descriptors_1, descriptors_2, matches );

DMatch类存放匹配结果

struct DMatch

{      

          int queryIdx;  //此匹配对应的查询图像的特征描述子索引

          int trainIdx;   //此匹配对应的训练(模板)图像的特征描述子索引

          int imgIdx;    //训练图像的索引(若有多个)

          float distance;  //两个特征向量之间的欧氏距离，越小表明匹配度越高。

          bool operator < (const DMatch &m) const;

};

匹配函数match (matches中保存着描述子之间的匹配关系)

    vector<DMatch> matches;

    BFMatcher bfmatcher(NORM_HAMMING);

    bfmatcher.match(descriptors1, descriptors2, matches);

 //-- 第四步:匹配点对筛选

1.计算最小距离和最大距离

找出所有匹配之间的最小距离和最大距离, 即是最相似的和最不相似的两组点之间的距离

    double min_dist=10000, max_dist=0;   //初始化

// 方法一

    for ( int i = 0; i < descriptors_1.rows; i++ )

    {

        double dist = matches[i].distance;//汉明距离matches[i]

        if ( dist < min_dist ) min_dist = dist;

        if ( dist > max_dist ) max_dist = dist;

    }   

  // 方法二   仅供娱乐的写法

    min_dist = min_element( matches.begin(), matches.end(), [](const DMatch& m1, const DMatch& m2) {return m1.distance<m2.distance;} )->distance;

    max_dist = max_element( matches.begin(), matches.end(), [](const DMatch& m1, const DMatch& m2) {return m1.distance<m2.distance;} )->distance;

    //方法三    更好的写法

  auto min_max = minmax_element(matches.begin(), matches.end(), [](const DMatch &m1, const DMatch &m2) { return m1.distance < m2.distance; });

  double min_dist = min_max.first->distance;

  double max_dist = min_max.second->distance;

    printf ( "-- Max dist : %f \n", max_dist );

    printf ( "-- Min dist : %f \n", min_dist );

2.去除错误匹配

    //当描述子之间的距离大于两倍的最小距离时,即认为匹配有误.但有时候最小距离会非常小,设置一个经验值30作为下限.

    std::vector< DMatch > good_matches;

    for ( int i = 0; i < descriptors_1.rows; i++ )

    {

        if ( matches[i].distance <= max ( 2*min_dist, 30.0 ) )

        {

            good_matches.push_back ( matches[i] );

        }

    }

 //-- 第五步:绘制匹配结果

    Mat img_match;//所有匹配

    Mat img_goodmatch;//好的匹配

    drawMatches ( img_1, keypoints_1, img_2, keypoints_2, matches, img_match );//画匹配线

    drawMatches ( img_1, keypoints_1, img_2, keypoints_2, good_matches, img_goodmatch );

    imshow ( "所有匹配点对", img_match );

    imshow ( "优化后匹配点对", img_goodmatch );

    waitKey(0);

    return 0;

}

/--------------------------------------------------------------------------------------------------------------------------------------------------------------------/

结果

特征点

所有匹配的点对

优化后匹配点对