之前写过一遍关于学习surf算法的blog:http://blog.csdn.net/sangni007/article/details/7482960

但是代码比较麻烦,而且其中还涉及到flann算法(其中的Random KDTree+KNN),虽然能看明白,但是比较费劲,今天在文档中找到一个简化版本:

1.SurfFeatureDetector detector( minHessian );构造surf检测器;

detector.detect( img_1, keypoints_1 ); detector.detect( img_2, keypoints_2 );检测

2.SurfDescriptorExtractor extractor;提取描述结构

Mat descriptors_1, descriptors_2;

extractor.compute( img_1, keypoints_1, descriptors_1 ); extractor.compute( img_2, keypoints_2, descriptors_2 );

3.BruteForceMatcher< L2<float> > matcher;牛逼的匹配结构啊!!!!可以直接暴力测量距离

std::vector< DMatch > matches;

matcher.match( descriptors_1, descriptors_2, matches );

文档:http://opencv.itseez.com/modules/gpu/doc/feature_detection_and_description.html?highlight=bruteforce#gpu::BruteForceMatcher_GPU

PS:OpenCV 你是在太强悍了!!!只有我想不到,木有你办不到的啊! 我真心跪了!

  1. /**
  2. * @file SURF_descriptor
  3. * @brief SURF detector + descritpor + BruteForce Matcher + drawing matches with OpenCV functions
  4. * @author A. Huaman
  5. */
  6. #include <stdio.h>
  7. #include <iostream>
  8. #include "opencv2/core/core.hpp"
  9. #include "opencv2/features2d/features2d.hpp"
  10. #include "opencv2/highgui/highgui.hpp"
  11. using namespace cv;
  12. using namespace std;
  13. void readme();
  14. /**
  15. * @function main
  16. * @brief Main function
  17. */
  18. int main( int argc, char** argv )
  19. {
  20. //if( argc != 3 )
  21. //{ return -1; }
  22. Mat img_1 = imread( "D:/src.jpg", CV_LOAD_IMAGE_GRAYSCALE );
  23. Mat img_2 = imread( "D:/Demo.jpg", CV_LOAD_IMAGE_GRAYSCALE );
  24. if( !img_1.data || !img_2.data )
  25. { return -1; }
  26. //-- Step 1: Detect the keypoints using SURF Detector
  27. int minHessian = 400;
  28. double t=getTickCount();
  29. SurfFeatureDetector detector( minHessian );
  30. std::vector<KeyPoint> keypoints_1, keypoints_2;
  31. detector.detect( img_1, keypoints_1 );
  32. detector.detect( img_2, keypoints_2 );
  33. //-- Step 2: Calculate descriptors (feature vectors)
  34. SurfDescriptorExtractor extractor;
  35. Mat descriptors_1, descriptors_2;
  36. extractor.compute( img_1, keypoints_1, descriptors_1 );
  37. extractor.compute( img_2, keypoints_2, descriptors_2 );
  38. //-- Step 3: Matching descriptor vectors with a brute force matcher
  39. BruteForceMatcher< L2<float> > matcher;
  40. std::vector< DMatch > matches;
  41. matcher.match( descriptors_1, descriptors_2, matches );
  42. t=getTickCount()-t;
  43. t=t*1000/getTickFrequency();
  44. //-- Draw matches
  45. Mat img_matches;
  46. drawMatches( img_1, keypoints_1, img_2, keypoints_2, matches, img_matches );
  47. cout<<"Cost Time:"<<t<<endl;
  48. //-- Show detected matches
  49. imshow("Matches", img_matches );
  50. waitKey(0);
  51. return 0;
  52. }
  53. /**
  54. * @function readme
  55. */
  56. void readme()
  57. { std::cout << " Usage: ./SURF_descriptor <img1> <img2>" << std::endl; }

学习OpenCV——Surf简化版-LMLPHP

图像中match的keypoints没有经过过滤。导致匹配点过多

文档地址:http://opencv.itseez.com/doc/tutorials/features2d/feature_description/feature_description.html?highlight=description

文档中还有一个版本带定位的和过滤Match的,

http://opencv.itseez.com/doc/tutorials/features2d/feature_homography/feature_homography.html?highlight=drawmatchesflags

from: http://blog.csdn.net/yangtrees/article/details/7544133

04-16 08:32