From: "Vladimir Z. Nuri" <vznuri@netcom.com>
Date: Wed, 9 Sep 1998 10:30:54 +0800
To: cypherpunks@cyberpass.net
Subject: IP: Machine Vision For Biometric Applications
Message-ID: <199809090224.TAA18267@netcom13.netcom.com>
MIME-Version: 1.0
Content-Type: text/plain




From: believer@telepath.com
Subject: IP: Machine Vision For Biometric Applications
Date: Wed, 09 Sep 1998 09:07:57 -0500
To: believer@telepath.com

----------------------
NOTE:  Document contains instructive images.  You should go to the website
and open the images.
----------------------

Source: Applied Optics Group at the University of Kent at Canterbury (U.K.)
http://speke.ukc.ac.uk/physical-sciences/aog/facereco/

Machine Vision For Biometric Applications 

We are currently conducting research in the areas of automated facial
recognition and data compression of digitised images of the human face.
This work began by performing an eigenfactor  analysis on a data set
comprising 290 faces drawn largely from the student population at the
University of Kent, Canterbury. 

The image below shows the first three components (eigenfaces) resulting
from this analysis. It is interesting to note that eigenfaces 2 and 3 have
a clear relationship to gender. Thus the addition of eigenface 2 to the
average eigenface (1) results in a feminine face whereas the subtraction of
face 2 produces a face having masculine characteristics. In a similar way,
addition of face 3 to the average produces a masculine face and subtraction
of face 3 from the average results in a face having feminine features. 

This approach, variously known as the Karhunen-Loeve expansion, eigenfactor
analysis, principal components or the Hotelling transform has exceptional
data compression properties when applied to this particular pattern class
(2-D images of human faces).

Below, we show the image reconstruction quality that is achievable using
codes of varying lengths which describe how to reconstitute the image using
the component eigenfaces as a basis. Note that the subject shown here was
not included in the original data set used to generate the eigenface basis.
Despite this, recognition is achieved using a very short code. 

This method works particularly well when conditions such as head-camera
orientation and subject illumination are controlled.We are now
investigating other methods (some related to the Karhunen-Loeve expansion,
some not) which may be suitable for automated facial recognition under less
benign conditions. In particular, we are beginning to investigate the use
of illumination compensation techniques and 3-D imaging techniques which
are independent of illumination conditions. 

The group working on facial recognition here at UKC collaborates with a
number of commercial/industrial organisations in the U.K. and Europe. We
currently await the outcome of a cooperative research bid (CRAFT) to the
European Commission which will involve the development of facial biometrics
for smart cards and other access control applications. The industrial
partners are Neural Computer Sciences (U.K.), Datastripe Ltd (U.K.), Inside
Technologies (France), Smartkort (Iceland) and A la Carte (Belgium). 

More links on Facial Recognition 

 Staff Involved 

 Dr. C.J. Solomon - E-mail: C.J.Solomon@ukc.ac.uk 
 Jamie P. Brooker - E-mail: jpb3@ukc.ac.uk 
-----------------------
NOTE: In accordance with Title 17 U.S.C. section 107, this material is
distributed without profit or payment to those who have expressed a prior
interest in receiving this information for non-profit research and
educational purposes only. For more information go to:
http://www.law.cornell.edu/uscode/17/107.shtml
-----------------------




**********************************************
To subscribe or unsubscribe, email:
     majordomo@majordomo.pobox.com
with the message:
     (un)subscribe ignition-point email@address
**********************************************
www.telepath.com/believer
**********************************************