Original publication


Main steps of the algorithm

  1. Detect keypoint candidates
    1. DoG is used, which is an approximation of the LoG operator
  2. Subpixel-localization of keypoints, discarding low-contrast + edge keypoints
  3. Compute dominant gradient orientation (weighted by gradient magnitudes)
  4. Compute descriptor (4×4 cells, 8 orientation bins =⇒ 16*8=128 scalar values)

This slide of a talk by Jason Clemens provides a good overview on the main steps of the SIFT algorithm:

Best introductions


Short & excellent introduction

What is the DOG threshold? What is the edge threshold?

  • here is an explanation with two examples images showing the meaning of both parameters
  • here is another explanation which points into another direction concerning the meaning of the edge threshold, namely that it is good for removing keypoints on edges
public/sift.txt · Last modified: 2014/01/06 14:07 (external edit) · []
Recent changes RSS feed Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki