Brute Force Matching with L2-Norm

asked 2017-10-25 07:48:00 -0600

Grillteller
121 ●1 ●10

When I use the default method L2-Norm in cv::BFMatcher on an own defined descriptor, is it irrelevant which position every value in my descriptor vector has?

So when I have a descriptor vector (1, 1, 2, 2, 6, 8, 4, 5, ...) according to http://mathworld.wolfram.com/L2-Norm.... it should have the same L2-Norm as (6, 2, 5, 4, 2, 8, 2, 1, 1...).

When I would use NORM_HAMMING is the position then relevant? Am I right or completely confused?

edit retag flag offensive close merge delete

add a comment

1 answer

Sort by » oldest newest most voted

answered 2017-10-25 09:50:48 -0600

berak
32993 ●7 ●81 ●312

updated 2017-10-25 10:15:22 -0600

for both L2 and HAMMING norm, the position of the values is irrelevant, as long as you stay consistant with the ordering

 (x1*x2 + y1*y2 + z1*z2) == (y1*y2 + x1*x2 + z1*z2) == (z1*z2 + y1*y2 + x1*x2) == ...

let me try another drawing:

// ok.
(x1 y1 z1)
(x2 y2 z2)

// still ok (same result as above):
(z1 y1 x1)
(z2 y2 x2)

// not ok:
(x1 y1 z1)
(z2 y2 x2)

your example above will "fail" for both norms, since you arbitrarily change the order in one of the vectors. again, if you use the same permutation for all of your vectors, the results will be the same.