We have shown results here for an accurate iris recognition
system, the methodology being a feature extraction
application in the eigen space domain. The rectangular iris
codes obtained from eye images are converted to iris vectors,
and from the set of iris vectors of a person a set of eigen iris
vectors are calculated. We considered different no of eigen iris
vectors, 10, 7, and 4 to evaluate the performance of the
recognition system. Also we have considered different cases
where a given iris code may be noisy, compressed, and
compressed with noise. In all cases we find that the technique
accurately recognizes the person when used with 10 and 7 nos
of eigen iris vectors, the difference being in the order of
values of the Euclidean distances. While using 4 nos of eigen
iris vectors for each group of person, we see that performance
is not so convincing for the third case. The technique is found
to give accurate and convincing results even with partial iris
codes of size 120x200 and also when some portion of the iris
is missing. To the best of our knowledge this is the first report
of iris recognition system using feature extraction in lower
dimensionality using eigen spaces.