OpenCV SIFT特征学习 (三) 特征如何更好的匹配（上）

BruteForceMatcher与FlannBasedMatcher

OpenCV自带的匹配方式主要有两种，蛮力法 BruteForceMatcher与 FlannBasedMatcher

BruteForceMatcher

        匹配时，假设输入两组kyepoints为kp1,kp2，则按顺序kp1的点与kp2中的匹配，距离最近的即为匹配点，故匹配结果是产生数量与kp1相同的DMatch。距离的计算有以下方法：L1（ 绝对值相加，又称曼哈顿距离 ）、L2（欧几里得距离）、Hamming、HammingLUT

FlannBasedMatcher

        首先要理解Flann：FLANN（ Fast Approximate Nearest Neighbor Search Library，快速近似最近邻搜索库   ），是一个快速最近邻搜索优化算法的集合，用于大数据集以及高维数据的检索。介绍详见 http://docs.opencv.org/modules/flann/doc/flann_fast_approximate_nearest_neighbor_search.html

​       代码如下（参考 http://docs.opencv.org/doc/tutorials/features2d/feature_flann_matcher/feature_flann_matcher.html ）：

  
  

   
   

    
    	cv::Mat image1=cv::imread("D:\\sift1.bmp");
   
   
   
   

    
    	cv::Mat image2=cv::imread("D:\\sift2.bmp");
   
   
   
   

    
    	cv::Mat out;
   
   
   
   

    
    	cv::SiftFeatureDetector  siftdtc; 
   
   
   
   

    
        std::vector<cv::KeyPoint> kp1,kp2;
   
   
   
   

    
        siftdtc.detect(image1,kp1);
   
   
   
   

    
    	siftdtc.detect(image2,kp2);
   
   
   
   

    
    	
   
   
   
   

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

    
    	SiftDescriptorExtractor extractor;
   
   
   
   

    
    	Mat descriptor1,descriptor2;
   
   
   
   

    
    	Mat img_matches;
   
   
   
   

    
    	extractor.compute(image1,kp1,descriptor1);
   
   
   
   

    
    	extractor.compute(image2,kp2,descriptor2);
   
   
   
   

    
     
   
   
   
   

    
    	//-- Step 3: Matching descriptor vectors using FLANN matcher
   
   
   
   

    
    	FlannBasedMatcher matcher;
   
   
   
   

    
    	std::vector< DMatch > matches;
   
   
   
   

    
    	matcher.match( descriptor1, descriptor2, matches );
   
   
   
   

    
     
   
   
   
   

    
    	double max_dist = 0; double min_dist = 100;
   
   
   
   

    
     
   
   
   
   

    
    	//-- Quick calculation of max and min distances between keypoints
   
   
   
   

    
    	for( int i = 0; i < descriptor1.rows; i++ )
   
   
   
   

    
    	{ double dist = matches[i].distance;
   
   
   
   

    
    	if( dist < min_dist ) min_dist = dist;
   
   
   
   

    
    	if( dist > max_dist ) max_dist = dist;
   
   
   
   

    
    	}
   
   
   
   

    
     
   
   
   
   

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

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

    
     
   
   
   
   

    
    	//-- Draw only "good" matches (i.e. whose distance is less than 2*min_dist,
   
   
   
   

    
    	//-- or a small arbitary value ( 0.02 ) in the event that min_dist is very
   
   
   
   

    
    	//-- small)
   
   
   
   

    
    	//-- PS.- radiusMatch can also be used here.
   
   
   
   

    
    	std::vector< DMatch > good_matches;
   
   
   
   

    
     
   
   
   
   

    
    	for( int i = 0; i < descriptor1.rows; i++ )
   
   
   
   

    
    	{ if( matches[i].distance <= max(2*min_dist, 0.02) )
   
   
   
   

    
    	{ good_matches.push_back( matches[i]); }
   
   
   
   

    
    	}
   
   
   
   

    
     
   
   
   
   

    
    	//-- Draw only "good" matches
   
   
   
   

    
    //	Mat img_matches;
   
   
   
   

    
    	drawMatches( image1, kp1, image2, kp2,
   
   
   
   

    
    		good_matches, img_matches, Scalar::all(-1), Scalar::all(-1),
   
   
   
   

    
    		vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );
   
   
   
   

    
     
   
   
   
   

    
    	//-- Show detected matches
   
   
   
   

    
    	imshow( "Good Matches", img_matches );
   
   
   
   

    
     
   
   
   
   

    
    	for( int i = 0; i < (int)good_matches.size(); i++ )
   
   
   
   

    
    	{ printf( "-- Good Match [%d] Keypoint 1: %d  -- Keypoint 2: %d  \n", i, good_matches[i].queryIdx, good_matches[i].trainIdx ); }
   
   
   
   

    
    	cv
    
    ::
    
    waitKey
    
    (
    
    0
    
    );
    
          
   
   
  
  

由上图得，仍存在不少同名匹配点（即kp1与kp2不是1对1的关系 ），那么如何更好的筛选呢？常用的方法有随记采样一致算法（RANSAC）,详情请看下一章。