Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!uw-beaver!uw-entropy!brothers!adrianb
From: adrianb@brothers.stat.washington.edu (Adrian Baddeley)
Newsgroups: comp.graphics
Subject: Image metrics
Message-ID: <2190@uw-entropy.ms.washington.edu>
Date: 18 Aug 89 22:51:25 GMT
Sender: news@uw-entropy.ms.washington.edu
Reply-To: adrianb@castor.ms.washington.edu
Distribution: usa
Organization: (visitor)UW Statistics, Seattle
Lines: 22

Since we're all talking about the accuracy of fractal compressions etc.,
are there any good references on the question of 

	measuring the discrepancy between two images?

In practice I've only ever seen this done with an L^p metric, 
e.g.
	- number of pixels that have different values in the two images
	- sum of absolute differences between corresponding pixel values
	- root-mean-square "	"	"	"	"	"	"

The mathematics of computed tomography is often done with a Sobolev
metric, somewhat harder to describe, but this is not really appropriate
to the present problem.

Comments?

----
adrianb@castor.ms.washington.edu	(until 21 august 1989)
Adrian Baddeley, visiting Department of Statistics GN-22,
University of Washington, Seattle WA 98195, USA.
	tel (U of W): +1 206 545-2617 / 543-7237