OpenCV基于DLCO描述子匹配

一：局部特征描述子介绍

2014年VGG发表了一篇基于凸优化的局部特征描述子学习(DLCO)的论文，OpenCV3.2以后在扩展模块中对该论文的完成了代码实现并发布了API支持，提供了基于DLCO的描述子生成支持、基于生成的描述子，可以实现图像特征匹配的对象识别。关于特征描述子学习相关的细节可以看这里：

http://www.robots.ox.ac.uk/~vgg/software/learn_desc/

提供了描述子模型，学习数据，C++版本实现的源代码下载

二：OpenCV程序演示

OpenCV中VGG的DLCO描述子生成支持下面几种

• VGG_120 = 100,

• VGG_80 = 101,

• VGG_64 = 102,

• VGG_48 = 103

默认支持输出描述子是120个向量即VGG_120。基于DLCO在OpenCV中代码实现对象检测与匹配大致分为如下几步：

1.加载图像

1. Mat box = imread("D:/vcprojects/images/box.png");

2. Mat scene = imread("D:/vcprojects/images/box_in_scene.png");

3. imshow("box image", box);

4. imshow("scene image", scene);

2.关键点检测（SURF）

1. Ptr<SURF> detector = SURF::create();

2. int minHessian = 400;

3. vector<KeyPoint> keypoints_1, keypoints_2;

4. detector->setHessianThreshold(minHessian);

5. detector->detect(box, keypoints_1);

6. detector->detect(box_scene, keypoints_2);

3.描述子生成（DLCO）

1. Ptr<VGG> vgg_descriptor = VGG::create();

2. Mat descriptors_1, descriptors_2;

3. vgg_descriptor->compute(box,  keypoints_1, descriptors_1);

4. vgg_descriptor->compute(box_scene, keypoints_2, descriptors_2);

4.特征匹配实现对象识别

1. // 计算匹配点

2. FlannBasedMatcher matcher;

3. std::vector< DMatch > matches;

4. matcher.match(descriptors_1, descriptors_2, matches);

5. double max_dist = 0; double min_dist = 100;

6. // 计算最大与最小距离

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

8. {

9.    double dist = matches[i].distance;

10.    if (dist < min_dist) min_dist = dist;

11.    if (dist > max_dist) max_dist = dist;

12. }

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

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

15. // 寻找最佳匹配，距离越小越好

16. std::vector< DMatch > good_matches;

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

18. {

19.    if (matches[i].distance <= min(2 * min_dist, 1.5))

20.    {

21.        good_matches.push_back(matches[i]);

22.    }

23. }

24. // 绘制最终匹配点

25. Mat img_matches;

26. drawMatches(box, keypoints_1, box_scene, keypoints_2,

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

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

29. //-- Localize the object

30. std::vector<Point2f> obj;

31. std::vector<Point2f> scene;

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

33. {

34.    //-- Get the keypoints from the good matches

35.    obj.push_back(keypoints_1[good_matches[i].queryIdx].pt);

36.    scene.push_back(keypoints_2[good_matches[i].trainIdx].pt);

37. }

38. Mat H = findHomography(obj, scene, RANSAC);

39. //-- Get the corners from the image_1 ( the object to be "detected" )

40. std::vector<Point2f> obj_corners(4);

41. obj_corners[0] = cvPoint(0, 0); obj_corners[1] = cvPoint(box.cols, 0);

42. obj_corners[2] = cvPoint(box.cols, box.rows); obj_corners[3] = cvPoint(0, box.rows);

43. std::vector<Point2f> scene_corners(4);

44. perspectiveTransform(obj_corners, scene_corners, H);

45. //-- Draw lines between the corners (the mapped object in the scene - image_2 )

46. line(img_matches, scene_corners[0] + Point2f(box.cols, 0), scene_corners[1] + Point2f(box.cols, 0), Scalar(0, 255, 0), 4);

47. line(img_matches, scene_corners[1] + Point2f(box.cols, 0), scene_corners[2] + Point2f(box.cols, 0), Scalar(0, 255, 0), 4);

48. line(img_matches, scene_corners[2] + Point2f(box.cols, 0), scene_corners[3] + Point2f(box.cols, 0), Scalar(0, 255, 0), 4);

49. line(img_matches, scene_corners[3] + Point2f(box.cols, 0), scene_corners[0] + Point2f(box.cols, 0), Scalar(0, 255, 0), 4);

50. //-- Show detected matches

51. imshow("Good Matches & Object detection", img_matches);

原图：

特征匹配结果

更多相关文章阅读

知不足者好学，

耻下问者自满！

关注【OpenCV学堂】

长按或者扫码下面二维码即可关注

OpenCV深度学习 

+群 573300093