Mirror Image

Mostly AR and Stuff

Oriented descriptors vs upright

I have tested oriented descriptors SURF descriptors vs upright descriptors for approximately horizontally oriented camera images and got feature density less than oriented then for upright. Repeatability of oriented was worse too…

17, March, 2009 - Posted by | Coding AR | , , ,


  1. Which implementation of SURF were you using? I’m the author of OpenSURF and after reading this I’m keen to do the tests myself!

    Comment by Chris | 21, March, 2010

  2. I have started with your own OpenSURF, but later did some modifications. Can’t remember exactly right now (was about a year ago), I think I replaced Hessian with boxed approximation of DoG, and also used heavily “padded”(overlapped) Haar cells in template. Nevertheless I tried quite a lot of different descriptors later, based on combinations of Haars, gradients and moments, and always upright descriptors were more useful for practical proposes, that is then translational movements dominate rotations.

    Comment by mirror2image | 22, March, 2010

Sorry, the comment form is closed at this time.

%d bloggers like this: