Re: fast curve similarity needed
- From: MCammarano <nobody@xxxxxxxxxxxx>
- Date: Fri, 07 Oct 2005 14:57:57 -0700
You might try comparing the distance matrices for the curves. For a cuve defined by N points, the distance matrix would be the NxN array of all the pairwise distances. The distance matrix is invariant under rotations of the curve, and by normalizing the distances you can make it invariant under uniform scaling as well. Note that it is also invariant for mirror-images, which may or may not be desirable for you.
You can probably find some papers about this technique being used to compare protein structures. This technique should be very fast given that you only have ~50 points per curve.
-Mike
Chris wrote:
Hello,
I'm looking for some algorithm (or preferably C++ code) to measure similarity between 3D curves (defined as a sequence of on the order of 50 3D points connected by lines) . The functions should have the option to match curves regardless of rotation in 3D space, and be preferably invariant in terms of uniform scale as well. The code needs to be as fast as possible; I'm willing to trade accuracy for speed. (say for the sake of argument several hundred curve-pair matches per second). Any ideas ? relevent papers ?
Thanks,
Chris
.
- Follow-Ups:
- Re: fast curve similarity needed
- From: Chris
- Re: fast curve similarity needed
- References:
- fast curve similarity needed
- From: Chris
- fast curve similarity needed
- Prev by Date: Re: fast curve similarity needed
- Next by Date: Sepia
- Previous by thread: Re: fast curve similarity needed
- Next by thread: Re: fast curve similarity needed
- Index(es):
Relevant Pages
|
