Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!usenet.eel.ufl.edu!news-feed-1.peachnet.edu!gatech!howland.reston.ans.net!news.sprintlink.net!EU.net!uknet!festival!leeds.ac.uk!news
From: stuart@dcre.leeds.ac.uk (S Butterfield (CoMIR))
Subject: Geometric Hashing
Message-ID: <1995Mar1.150824.18931@leeds.ac.uk>
Date: Wed, 1 Mar 1995 15:08:23 +0000 (GMT)
X-Newsreader: TIN [version 1.2 PL2]
Lines: 25

I have recently been looking into the use of Wolfson and Lamdan's
Geometric Hashing technique, for model based recognition. Although
I understand the theory, I'm having trouble devising a suitable
hash function, and none of the papers I have tracked down go into
such implementation-specific detail.

The problem is to create a hash function `H':

	i=H(a)

such that `H' returns a value in the range 0..i for all `a'.
In my case, the `a' values are real numbers in the range -10..+10,
and `i' is around 100,000. Ideally the hash function should give
an even distribution over `i' for all values of `a', but my attempts
so far have failed miserably, covering on average just 10% of the
possible `i' values.

I'd be extremely grateful for any suggestions people may have on
how to go about constructing such a function, or any pointers to
papers which I might not yet have seen.

Thanks in advance,

--
Stuart
