Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!gatech!howland.reston.ans.net!vixen.cso.uiuc.edu!sdd.hp.com!hp-pcd!hpcvsnz!briand
From: briand@cv.hp.com (Brian Dixon)
Subject: Re: Urgent: Help!
Sender: news@hpcvsnz.cv.hp.com (News )
Message-ID: <D84ow0.2Cw@hpcvsnz.cv.hp.com>
Date: Fri, 5 May 1995 23:21:35 GMT
References: <1995May5.063750.18040@uxmail.ust.hk>
Nntp-Posting-Host: hpcvsgen.cv.hp.com
Organization: Hewlett-Packard
X-Newsreader: TIN [version 1.1 PL9.4]
Lines: 33

Law Kwok Gay (cs_pats@ug.cs.ust.hk) wrote:
: Hi,
:    We are doing a grey level picture recognition project. But we find the
: image intensity affect the recognition rate much. Could anyone tell me 
: how to get rid(some algorithms) of the light intensity condition so that the 
: gray picture is insensitive under different lightcondition.

It depends on what way you are doing 'recognition'.  Some algorithms are
less sensitive to overall image intensity variation, while others are
extremely sensitive to it.  Just what approach are you taking?

For example, if you can use template matching, it will be fairly resilient
to illumination differences in your acquired images...especially if you
'trained' your model after carefully setting up the lighting and focus for
best contrast.  If you are doing pattern matching, e.g. edge-based or
by extracting so-called invariant classifiers, then your results will depend
highly on your chosen algorithm.  It's tough to answer your question without
first knowing something more than just "we're doing recognition."

Before I forget (almost did), have you tried acquiring an image of a
uniform plane under the same or similar lighting conditions?  If you have,
then you can usually subtract it from an acquired image to correct for
background lighting variation.  But be forewarned that you'll also lose
affective dynamic range by doing this.  Also, subtraction works if your
camera is logarithmic with a gamma of 1.0.  If your camera is linear then
you need to divide the image by the background image.

Good luck,
Brian

--
Brian Dixon, Machine Vision Engineer, Hewlett Packard (Corvallis, Oregon)
503-715-3143 (wk), briand@cv.hp.com (email). "Opinions & attitudes are mine!"
