0

I have been trying to detect some features of an image and another image which is created by rotating the other image by 90 degrees. My aim was to confirm that OrbFeatureDetector with BriefFeatureExtractor is actually rotation invariant. However, literally, I have 0 good match. Not even one combination of 4 point matches correspond to a transformation matrix. I wanted to know if it is actually possible that there could be no good matches between an image and it's rotated version.

I could post codes, but I am pretty sure I have no errors with my code, it's just there are 0 matches. And also I wouldnt want to bother you with raw code.

Thanks for ideas.

GOod day.

Ozum Safa
  • 1,458
  • 2
  • 15
  • 27

1 Answers1

1

According to part 4 of BRIEF: Binary Robust Independent Elementary Features the descriptor "...is not designed to be rotationally invariant ..." but "Nevertheless ... it tolerates small amounts of rotation". So yes, it's possible and very likely that you get no matches with a rotation of 90 degress.

AD-530
  • 1,059
  • 9
  • 9
  • Thank you for your answer. I think you are right. 90 degrees is an extreme rotation. All the gradients changes, if even they use gradients. Thank you I will give more thinking to the papers. Good day. – Ozum Safa Nov 02 '12 at 07:51