Mirror Image

Mostly AR and Stuff

Testing a new descriptor.

Trying a new descriptor, inspired by SURF and SIFT. Want to use gradient instead of Haar transforms of intensity, but with less dimensionality than SURF. Also don’t need rotation/scale invariance, because using incremental tracking.

20, November, 2009 - Posted by | Coding AR | , , , , ,


  1. Use the DAISY descriptor from the cvlab EPFL

    Comment by Pierre | 7, December, 2009

  2. Yep, I know of DAISY, but it’s unreasonably computationally expensive in my opinion. I tried different combinations of integrals of gradient, absolute values of gradients, histograms of gradients, haar-like features etc, and didn’t see considerable benefits of one or another for upright incremental descriptors.

    Comment by mirror2image | 7, December, 2009

  3. what sort of app are you planning to use ? daisy is for dense ops and can be unnecessary for sparse matching. however, you may want to play with its parameters for shorter desc lengths…

    Comment by hk | 9, January, 2010

  4. For sparce exactly, and incremental tracking. Daisy is overkill in that case.

    Comment by mirror2image | 9, January, 2010

Sorry, the comment form is closed at this time.

%d bloggers like this: