Theory behind Feature2D modul; substitute Descriptors/Detectors

asked 2017-10-17 11:33:46 -0500

Sanchopanza gravatar image

updated 2017-10-23 02:02:37 -0500

Hello everyone,

I have a small understanding problem with feature matching from the Feature2D-matching modul and I hope you can help me:

If I understand the theory about this topic in the right way, there seem to be detector which are invariant (or estimate) the scale and rotation of the intrest-points (as SIFT, ORB) and some of them do not (as FAST).

In the same way some descriptors do take this parameters into account (HOG, ORB), some do not: (BRIEF);

1.: Eg. if i combine a FAST-Detector with an BRIEF descriptor - no rotated or scaled Part will be found in the feature matching process, am i right?

2.: At what point will the estimation of an rotationangle take place? If a combination of a FAST-Detector with an SIFT-descriptor does descriptor estimate the strongest gradient first and normalizes the histogramm, or does it need the angle and if not provided like in FAST it is supposed to be zero?

3.: Just to be sure: Are SIFT descriptor and HOG the same thing?

edit retag flag offensive close merge delete