All geek questions in one place

Getting the most significant functions obtained from SIFT / SURF

I use SURF to extract functions from images and map them to others. My problem is that some images have over 20,000 functions that slow down when compared to crawl.

Is there a way to extract only the most important functions from this set?

I tried to calculate the MSER for the image and use only those functions that are in these regions. This gives me a decrease from 5% to 40%, without affecting the quality of the comparison negatively, but it is unreliable and still not enough.
I could additionally resize the image, but I seem to have greatly influenced the quality of the functions in some cases.
SURF offers several parameters (hessian threshold, octaves and layers per octave), but I could not find anything about how changing these parameters will affect the value of the function.

+3
source share
1 answer

, , , hessian, .
, n:

  • ,
  • , MSER.
  • , , hessian
  • n , , 75% - hessian.
  • , hessian,

n, 1500 .

+4

Source: https://habr.com/ru/post/1781250/

More articles:


All Articles