Robust Identification Based on a Processing-Oriented Object Representation

                         Georgii Khachaturov

             Universidad Autonoma Metropolitana, Mexico

            Home page http://www.tarunz.org/~xgeorge
                 or : http://newton.uam.mx/xgeorge

                              Abstract

There is a gap in most of pattern recognition approaches between
low-level processing and high-level analysis. A recent approach filled
this gap for the problem of structural identification: It allows
execution of the low-level image processing routines under full
control of an identification task.

A new object representation, called TSG-model, was developed for this
purpose. Specific property of the TSG-model is that its atomic
elements are related directly to the low-level processing routines,
unlike traditional approaches regarding so far geometrical concepts as
atomic elements of an object model.  TSG-model allows the master level
to decompose processing for an identification task into a stream of
verification of hypotheses. Controlling this stream, the master level
indicates at each step where, what, and how to process next.

The most recent development provides this approach with
robustness. Using simple analogy, one may say that now the TSG-model
works like a redundant code whereas the image processing looks like
error tolerant signal transmission .  So omissions of some object
details during detection as well partial occlusions are allowed now in
the framework of the approach.