0
$\begingroup$

I couldn't figure out below paragraph on SURF paper and hope that someone can help me to understand it.

Bay H., Ess A., Tuytelaars T. Van Gool L. - Speed-Up Robust Features (SURF), page 4, column 2

The large scale changes, especially between the first filter within these octaves (from 9 to 15 is a change of 1:7), renders the sampling of scales quite crude. Therefore, we have also implemented a scale space with a finer sampling of the scales. This first doubles the size of the image, using linear interpolation, and then starts the first octave by filtering with a filter of size 15. Additional filter sizes are 21, 27, 33, and 39. Then a second octave starts, again using filter which now increase their sizes by 12 pixels, after which a third and fourth octave follow. Now the scale change be- tween the first two filter is only 1.4 (21/15). The lowest scale for the accurate version that can be detected through quadratic interpolation is s = (1.2(18/9)/2)=1.2.

is that mean they used two different scale space at same time or they used each one individually and choose the one with better result?

$\endgroup$
-1
$\begingroup$

They implemented each one separately, then compared their results to each other.

Repeatibility scores

FH-9 ---> it's a detector that it's initial filter size is 9x9

FH-15 --> it's a detector that it's initial filter size is 15x15

$\endgroup$
  • $\begingroup$ i guess i should leave this question unanswered . too many down votes. i would like to know the reason if you don't mind. $\endgroup$ – Alireza Jun 27 '14 at 14:07
  • $\begingroup$ You are absolutely right. This may be considered as two different algorithms. First is using 9x9 filter and doesn't scale the image, the second one doubles the image and use 15x15 filter. I prefer the first one. $\endgroup$ – kelin Oct 14 '15 at 13:57

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.