Multi-channel
cooperativity in visual processing (MCCOOP)
NEST-2003-1 ADVENTURE,
Proposal/Contract no.:012963-2
Funding period:
2005-2008
Contact Person
Silvio.P. SABATINI
Department of Biophysical and Electronic Engineering
University of Genoa
Via Opera Pia, 11a
I-16145 Genova, ITALY
tel: (+39) 10 353 2092
fax: (+39) 10 353 2289
e-mail: silvio@dibe.unige.it
Fraunhofer Institute for Media Communication,
Katholieke
DIBE -
A computer with
intelligent visual skills comparable to biological ones has been an
ambitious
research goal ever since the field of computer vision took off some 30
years
ago. As an alternative to mathematical approaches, especially when they
fall
short of their expectations, attempts to mimic the processing of the
brain have
been pursued, but implementing the result into computer vision systems
is still
in its infancy.
This project
envisages the development of biologically-inspired computer vision
techniques
by exploiting the physiology and anatomical connectivity of individual
visual
neurons and integrating their individual functionalities into
cooperative visual
systems. Our approach will emphasize the role of the topological
organization
of visual neurons with similar functional properties. The novelty of
this
approach lies in the direct involvement of the neuron’s topological
organization ("spatial layout") for the processing of visual
information.
Two types of
visual neuron functionalities (``channels”) will be modelled and
integrated:
the orientation selectivity leading to selective edge detection and the
perception of 3-dimensional motion. These two functionalities will be
linked
together forming a cooperative multi-channel system for selective
object
detection under various environmental conditions. The capability of the
newly
developed computer vision techniques will be tested with images of
rapidly changing
scenes such as those observed by a car driver.
This project is in the rapidly evolving area of biologically-motivated computer vision, and at the cross road of computational neuroscience and computer vision. It has a high scientific risk since multi-channel modelling has been tried before, but failed to proliferate computer vision.
The project has a significant technological potential to contribute to the development of a new breed of biologically-motivated vision systems that segment and track objects in cluttered scenes.