Computer Assisted Visual Interactive Recognition (CAVIAR) Technology
Arthur Evans, John Sikorski, Patricia Thomas, Sung-Hyuk Cha, Charles Tappert
evansart@prodigy.net, john.sikorski@regeneron.com, pt84575w@pace.edu, scha@pace.edu,
ctappert@pace.edu
Pace University, White Plains, NY
Jie Zou, Abhishek Gattani, George Nagy
Rensselaer Polytechnic Institute, Troy, NY
zouj@alum.rpi.edu, gattani@gmx.net, nagy@ecse.rpi.edu
The distinctive aspect of the CAVIAR technology is a
visible, parameterized geometrical model that serves as
the human-computer communication channel. Evaluation
of CAVIAR flower and face recognition systems shows
that their accuracy is much higher than that of the
machine alone; their recognition time is much lower than
that of the human alone; they can be initialized with a
single reference sample per class; and they improve with
use. CAVIAR-flower has been ported to stand-alone and
to wireless laptop-client Personal Digital Assistants
(PDAs).
1. Introduction
We describe a progression of computer-assisted visual
interactive
recognition
(CAVIAR)
systems. Our
development was a Windows platform. We have ported
our system to a stand-alone Personal Digital Assistant,
and to a pocket PC configured as a wireless client to a
laptop, both with plug-in cameras. The key difference
between CAVIAR and most current classification
systems is operator interaction based on a visible model.
We report results on two very different applications,
flower classification and face recognition, and comment
on the merits of several recognition system architectures.
Automated visual recognition systems seldom achieve
100% correct classification on families of objects of
interest. Most allow user interaction at the beginning, to
locate or frame the object, and at the end, to classify
“rejects” or “low confidence” items [Heritaoglu01,
Zhang02]. Providing means of user interaction throughout
the process, rather than only at the beginning or end, is
more efficien