首页 > 代码库 > OpenCV2.4.4中调用SIFT特征检测器进行图像匹配

OpenCV2.4.4中调用SIFT特征检测器进行图像匹配

 

OpenCV中一些相关结构说明:

特征点类:

 class KeyPoint{        Point2f  pt;  //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; //        int  octave; //特征点所在的图像金字塔的组        int  class_id; //用于聚类的id}

存放匹配结果的结构:

 1 struct DMatch 2 { 3     //三个构造函数 4     DMatch(): queryIdx(-1), trainIdx(-1),imgIdx(-1), 5         distance(std::numeric_limits<float>::max()) {} 6     DMatch(int  _queryIdx, int  _trainIdx, float  _distance ) : 7         queryIdx( _queryIdx),trainIdx( _trainIdx), imgIdx(-1),distance( _distance) {} 8     DMatch(int  _queryIdx, int  _trainIdx, int  _imgIdx, float  _distance ) : 9         queryIdx(_queryIdx), trainIdx( _trainIdx), imgIdx( _imgIdx),distance( _distance) {}10 11     int queryIdx;  //此匹配对应的查询图像的特征描述子索引12     int trainIdx;   //此匹配对应的训练(模板)图像的特征描述子索引13     int imgIdx;    //训练图像的索引(若有多个)14     float distance;  //两个特征向量之间的欧氏距离,越小表明匹配度越高。15     bool operator < (const DMatch &m) const;16 };

 说明:以两个特征点描述子(特征向量)之间的欧氏距离作为特征点匹配的相似度准则,假设特征点对p和q的特征描述子分别为Desp和Desq,则其欧氏距离定义为:

 

所以每个匹配分别对应训练图像(train)和查询图像(query)中的一个特征描述子(特征向量)。

 1 #include "opencv2/highgui/highgui.hpp" 2 #include "opencv2/imgproc/imgproc.hpp" 3 #include "opencv2/nonfree/nonfree.hpp" 4 #include "opencv2/nonfree/features2d.hpp" 5 #include <iostream> 6 #include <stdio.h> 7 #include <stdlib.h> 8  9 using namespace cv;10 using namespace std;11 12 int main()13 {14     initModule_nonfree();//初始化模块,使用SIFT或SURF时用到15     Ptr<FeatureDetector> detector = FeatureDetector::create( "SIFT" );//创建SIFT特征检测器16     Ptr<DescriptorExtractor> descriptor_extractor = DescriptorExtractor::create( "SIFT" );//创建特征向量生成器17     Ptr<DescriptorMatcher> descriptor_matcher = DescriptorMatcher::create( "BruteForce" );//创建特征匹配器18     if( detector.empty() || descriptor_extractor.empty() )19         cout<<"fail to create detector!";20 21     //读入图像22     Mat img1 = imread("desk.jpg");23     Mat img2 = imread("desk_glue.jpg");24 25     //特征点检测26     double t = getTickCount();//当前滴答数27     vector<KeyPoint> keypoints1,keypoints2;28     detector->detect( img1, keypoints1 );//检测img1中的SIFT特征点,存储到keypoints1中29     detector->detect( img2, keypoints2 );30     cout<<"图像1特征点个数:"<<keypoints1.size()<<endl;31     cout<<"图像2特征点个数:"<<keypoints2.size()<<endl;32 33     //根据特征点计算特征描述子矩阵,即特征向量矩阵34     Mat descriptors1,descriptors2;35     descriptor_extractor->compute( img1, keypoints1, descriptors1 );36     descriptor_extractor->compute( img2, keypoints2, descriptors2 );37     t = ((double)getTickCount() - t)/getTickFrequency();38     cout<<"SIFT算法用时:"<<t<<""<<endl;39 40 41     cout<<"图像1特征描述矩阵大小:"<<descriptors1.size()42         <<",特征向量个数:"<<descriptors1.rows<<",维数:"<<descriptors1.cols<<endl;43     cout<<"图像2特征描述矩阵大小:"<<descriptors2.size()44         <<",特征向量个数:"<<descriptors2.rows<<",维数:"<<descriptors2.cols<<endl;45 46     //画出特征点47     Mat img_keypoints1,img_keypoints2;48     drawKeypoints(img1,keypoints1,img_keypoints1,Scalar::all(-1),0);49     drawKeypoints(img2,keypoints2,img_keypoints2,Scalar::all(-1),0);50     //imshow("Src1",img_keypoints1);51     //imshow("Src2",img_keypoints2);52 53     //特征匹配54     vector<DMatch> matches;//匹配结果55     descriptor_matcher->match( descriptors1, descriptors2, matches );//匹配两个图像的特征矩阵56     cout<<"Match个数:"<<matches.size()<<endl;57 58     //计算匹配结果中距离的最大和最小值59     //距离是指两个特征向量间的欧式距离,表明两个特征的差异,值越小表明两个特征点越接近60     double max_dist = 0;61     double min_dist = 100;62     for(int i=0; i<matches.size(); i++)63     {64         double dist = matches[i].distance;65         if(dist < min_dist) min_dist = dist;66         if(dist > max_dist) max_dist = dist;67     }68     cout<<"最大距离:"<<max_dist<<endl;69     cout<<"最小距离:"<<min_dist<<endl;70 71     //筛选出较好的匹配点72     vector<DMatch> goodMatches;73     for(int i=0; i<matches.size(); i++)74     {75         if(matches[i].distance < 0.31 * max_dist)76         {77             goodMatches.push_back(matches[i]);78         }79     }80     cout<<"goodMatch个数:"<<goodMatches.size()<<endl;81 82     //画出匹配结果83     Mat img_matches;84     //红色连接的是匹配的特征点对,绿色是未匹配的特征点85     drawMatches(img1,keypoints1,img2,keypoints2,goodMatches,img_matches,86                 Scalar::all(-1)/*CV_RGB(255,0,0)*/,CV_RGB(0,255,0),Mat(),2);87 88     imshow("MatchSIFT",img_matches);89     waitKey(0);90     return 0;91 }


结果:

 

效果图:

 

源码下载:

http://download.csdn.net/detail/masikkk/5511831

当然,这些匹配还没有经过系统的筛选,还存在大量的错配,关于匹配的筛选参见这篇文章:

 利用RANSAC算法筛选SIFT特征匹配

以及RobHess的SIFT源码分析系列文章:http://blog.csdn.net/masibuaa/article/details/9191309

 

OpenCV2.4.4中调用SIFT特征检测器进行图像匹配