"Rich Ulrich" <[EMAIL PROTECTED]> schreef in bericht
news:[EMAIL PROTECTED]
> On Tue, 02 Sep 2003 16:40:29 +0200, Cristian-Augustin Saita
> <[EMAIL PROTECTED]> wrote:
>
> > Hello,
> >
> > Someone could tell me, please, if there exists any standard method
of
> > generating points in a multi-dimensional real hyper-space (like
[0,1]^N,
> > where N represents the number of dimensions) so to preserve the
unknown
> > space distribution of a given sample of points from the considered
space.
> >
> > More precisely, based on a sample of 60,000 mutidimensional points
> > (representing colour characteristics of images from an image
database),
> > is it possible to generate a set of 1,000,000 points with
garanties (of
> > any nature) that the generated set preserves the (unknown) space
> > distribution of the original sample set? I want to avoid to simply
> > replicate/duplicate the original points, method which I'm not very
sure
> > that preserves the original distribution either.
>
> You can draw from a collection of the original points.
>
> Or you can use the original points but make them blurry.
>
> Or you can derive parameters from the original points
> and 'randomize'  based on the parameters.
>
> Once you have said "No" those three things, alone and in
> every combination,  I can't think of anything to suggest,
> except that you need to reconsider what you are trying
> to achieve.
>

Maybe you are looking for something like a "convex hull". Searching
with Google on these keywords results in a lot of usefull references.

Jos Jansen

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to