Given a database of a number of images of a single object (face,
train, etc.) photographed from different angles, and given 1 image of
the same object which is not in the database, I would like to find the
closest identified image in the database. Closeness is to be measured
in the L-2 norm of the difference in coefficients of the database
image and the probe image in the eigen basis.

The database of images I have are 30 greyscale *.pgm images of a face,
centered, but facing different directions.

I would like some Matlab code, or perhaps a pseudocode algorithm to
help me solve this problem. An answer before the 16th will recieve a
$5.00 tip.

---

Clarification of Question by garbonzo-ga on 09 Apr 2004 11:31 PDT

Also, I would like to show at least 3 dominant eigen images of any two
images in the database. (Eigen images are the images which are formed
by projecting the image onto the eigen vectors corresponding to the
largest eigen values.

Hi, garbonzo-ga:

A more general approach with images is to use singular values rather
than eigenvalues.  The effective identification of images is a
difficult computational problem.  If this is a "classroom" exercise in
which you are to learn the mechanics of numerical linear algebra
applications, I hope you will find it an interesting challenge.

regards, mathtalk-ga