Search code examples
siftvlfeat

default vlfeat sift return less features than sift implementation of D.lowe

As I mentioned in title default vlfeat sift returns less features than sift of D.Lowe. How could I let vlfeat sift return as much as features as D.Lowe's does. Or can we only offer position info of key-points to extract sift features with vlfeat sift? Thank you.

Solution

  • under deault settings, vlfeat sift return less features implementation of D.Lowe. The document of vlfeat point out:

    first octave index: set to -1 to extract very small features

    number of scale: can affect the number of extracted keypoints

    edge threshold: decrease to eliminate more keypoints

    peak threshold: increase to eliminate more keypoints

    So by changing the first octave index and number of scale can increase the number of returned feature.

    matlab code example:

    [f, d] = vl_sift(img, 'levels', 4, 'firstoctave', -1);