如何从SIFT算法产生的描述符中获取图像的关键点或像素坐标

Question

I have used SIFT algorithm to produce descriptor. After doing some processing to descriptor I got some selected descriptor. Now I want to draw lines between those descriptors, for that I want key points or pixel coordinates back from those descriptors.

I am trying to implement copy move forgery detection using opencv. Using key point concept, I got matched pairs of descriptor. Just want to draw line between those matched descriptor.

Code for producing descriptor:

sift = cv2.xfeatures2d.SIFT_create()
kp2, desc2 ift.detectAndCompute(gtemp,None)

The final descriptor I got, between which I want to draw lines:

[ 1. 64. 131. 1. 2. 9. 1. 0. 4. 80. 131. 0. 0. 3. 14. 12. 1. 52. 131. 0. 0. 0. 7. 9. 2. 25. 131. 0. 0. 0. 0. 0. 19. 12. 20. 3. 48. 54. 1. 8. 92. 28. 38. 0. 0. 2. 58. 108. 8. 18. 63. 0. 0. 43. 131. 28. 15. 18. 65. 0. 0. 31. 115. 33. 33. 2. 3. 47. 61. 8. 0. 7. 131. 24. 0. 0. 1. 0. 12. 47. 77. 4. 0. 0. 0. 27. 91. 22. 1. 0. 0. 0. 0. 28. 113. 19. 10. 0. 2. 9. 1. 2. 79. 55. 131. 1. 0. 0. 0. 0. 17. 131. 47. 0. 0. 0. 0. 0. 1. 25. 0. 0. 0. 0. 0. 0. 3. 1.] [ 23. 22. 31. 4. 0. 1. 8. 8. 26. 41. 40. 25. 4. 2. 13. 19. 11. 42. 46. 48. 8. 17. 15. 7. 4. 25. 28. 42. 38. 74. 34. 2. 68. 21. 24. 25. 2. 10. 11. 14. 14. 17. 119. 50. 12. 15. 11. 12. 87. 35. 36. 21. 14. 10. 7. 39. 8. 5. 14. 72. 119. 25. 3. 6. 5. 11. 13. 6. 1. 61. 113. 21. 8. 4. 7. 8. 9. 93. 119. 44. 69. 16. 13. 13. 15. 28. 45. 59. 5. 3. 7. 119. 119. 7. 0. 0. 0. 119. 119. 0. 0. 10. 28. 3. 0. 100. 119. 0. 0. 29. 54. 6. 0. 20. 55. 36. 52. 62. 24. 3. 2. 0. 4. 119. 119. 8. 1. 1.]

I think if I could get pixel coordinates back from this descriptors then it will be easier to draw lines between them. Using python 3.6 with OpenCV.

Answer 1

I understand that you want to draw lines between the two images that you are matching.
For this purpose OpenCV has cv.drawMatches . The documentation for this method is here .

Use it like this image3 = cv2.drawMatches(image1, kp1, image2, kp2, matches, None, flags=2) .

Also if you need pixel coordinates for your matched keypoints you have to read the pt attribute for cv2.KeyPoint .

For example if you want the pixel coordinates of all you matched features in image 2:

for kp in kp2:
    for match in matches:
        print(kp2[match.trainIdx].pt)

In this case match.trainIdx is the id for a match in reference of image2 and the pt attribute refers to the pixel coordinate of that feature point. The same goes for image1 but you have to read kp1[match.queryIdx].pt .