Will using Cosine-Similarity or Histogram Intersection as distance-metric, for matching SIFT descriptors, help to get better results?

Anybody tried on that ever?

  • $\begingroup$ I haven`t tried Cosine-Similarity or Histogram Intersection yet, but Arandjelovic has proposed to use Hellinger metric instead of L2 and shown that it gives better results. robots.ox.ac.uk/~vgg/publications/2012/Arandjelovic12/… $\endgroup$ – old-ufo Feb 16 '14 at 10:06
  • $\begingroup$ have you tried seas.upenn.edu/~ofirpele/SiftDist $\endgroup$ – object recognition Feb 16 '14 at 11:01
  • $\begingroup$ I have seen it, but since it cannot be used in kd-tree, it doesn`t fit me - n^2 matching is too slow. RootSIFT - instead - could be easily used in kd-tree. $\endgroup$ – old-ufo Feb 16 '14 at 11:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.