Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!howland.reston.ans.net!ix.netcom.com!netcom.com!perry
From: perry@netcom.com (Perry West)
Subject: Re: Resolution
Message-ID: <perryDAu6D7.7w5@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
X-Newsreader: TIN [version 1.2 PL1]
References: <3sh35c$2f1e@usenetw1.news.prodigy.com> <perryDAs6w7.CsL@netcom.com> <1995Jun26.174916.19831@frontier.tno.nl>
Date: Tue, 27 Jun 1995 14:46:18 GMT
Lines: 35
Sender: perry@netcom13.netcom.com

: > perry@netcom.com (Perry West) wrote:

<snip>

: This answer (of three I read) corresponds the most with my ideas about
: resolution. But a better description of "resolution in the view of

<snip>

: I do not agree fully with the use of the word "information" as used
: by Perry. Quite often with image processing, there is a model into
: which we try to fit an acquired image. If this model is correct, we
: can use characteristics of this model to extract non-acquired
: information from an image.

You are quite right.  When we combine the information in an image with 
a-priori information, we can get far more detailed information.  This is 
the basis for sub-pixel information.  But there is a very substantial risk.

: Example 1: The worst image of a ring/circle has only to contain three
: good parts to reconstruct the radius and position of this ring.

Here you are right if you are absolutely certain the object is a 
ring/circle.  As an example, a common technique is to do a least squares 
fit of edge points to a model; let's say the model is a straight line.  
If our model fits the real world, or rather if the real world fits our 
model, we can achieve a good deal of improvement in precision.  However, 
I have frequently run into the situation where an object was expected to 
have straight sides, but didn't.  Then the result of fitting the edge 
points to a straight line can actually be a degredation in the precision 
(resolution if you will) of the answer.  The error comes from using the 
model to assume information in the image that actually isn't there.

<snip>

