Feature matching - Minimum numbers of matches

asked 2015-10-28 16:28:46 -0500

I'm playing with "feature matching" in OpenCV. For different types of matching methods, we need to define a condition (i.e. OpenCV – SIFT SURF ORB tutorial MIN_MATCH_COUNT = 10) regarding the minimum number of matches.

I'm wondering if there is any kind of standard or golden standard method or criteria for setting this condition? How can / should we technically get this sort of threshold or condition?

Even a rule of thumb suggested would be help.

Thank you so much,

edit retag flag offensive close merge delete


This minimum is not for matching features but for homography. To find homography matrix H you need a minimum of 4 points (equivalent to 4 matching).

If you have more matching it's better (if they are good!)...

LBerger gravatar imageLBerger ( 2015-10-29 03:35:35 -0500 )edit

Thanks for your feedback. However, I have three objects and two of them are similar and one of them is different. I have more than 4 points matched between AB and AC. I'm wondering if there is any decision maker and algorithmic approach to find the best match ?

samansarraf gravatar imagesamansarraf ( 2015-10-29 18:36:51 -0500 )edit

you may use option crossCheck in BFMatching to improve quality and use distance value in DMatch too

LBerger gravatar imageLBerger ( 2015-10-30 02:12:45 -0500 )edit