Neuromorphic Attentional Selection for Efficient Allocation of Computing Resources. Itti, L. In Proc. Virtual Worlds and Simulation Conference, San Antonio, Texas, Jan, 2002.
abstract   bibtex   
When confronted with cluttered natural environments, animals still perform orders of magnitude better than artificial vision systems in tasks such as orienting, target detection, navigation and scene understanding. To better understand biological visual processing, we have developed a neuromorphic model of how our visual attention is attracted towards conspicuous locations in a visual scene. It replicates processing in the dorsal ("where") visual stream in the primate brain. The model includes a bottom-up (image-based) computation of low-level color, intensity, orientation and motion features, as well as a non-linear spatial competition that enhances salient locations in each feature channel. All feature channels feed into a unique scalar "saliency map" which controls where to next focus attention. We show how our simple within-feature competition for salience effectively suppresses strong but spatially widespread feature responses due to clutter. The model robustly detects salient targets in live outdoors video streams, despite large variations in illumination, clutter, and rapid egomotion. The success of this approach suggests that neuromorphic vision algorithms may prove unusually robust for outdoors vision applications. Further, we argue that the massively parallel attentional selection implemented in our model may represent an efficient approach to the general problem of allocating limited computational resources under conditions of sensory overload.
@inproceedings{ Itti02vwsim,
  author = {L. Itti},
  title = {Neuromorphic Attentional Selection for Efficient Allocation of Computing Resources},
  year = {2002},
  abstract = {When confronted with cluttered natural environments, animals
still perform orders of magnitude better than artificial vision
systems in tasks such as orienting, target detection, navigation and
scene understanding. To better understand biological visual
processing, we have developed a neuromorphic model of how our visual
attention is attracted towards conspicuous locations in a visual
scene.  It replicates processing in the dorsal ("where") visual
stream in the primate brain. The model includes a bottom-up
(image-based) computation of low-level color, intensity, orientation
and motion features, as well as a non-linear spatial competition that
enhances salient locations in each feature channel.  All feature
channels feed into a unique scalar "saliency map" which controls
where to next focus attention. We show how our simple within-feature
competition for salience effectively suppresses strong but spatially
widespread feature responses due to clutter.  The model robustly
detects salient targets in live outdoors video streams, despite large
variations in illumination, clutter, and rapid egomotion. The success
of this approach suggests that neuromorphic vision algorithms may
prove unusually robust for outdoors vision applications. Further, we
argue that the massively parallel attentional selection implemented in
our model may represent an efficient approach to the general problem
of allocating limited computational resources under conditions of
sensory overload.},
  booktitle = {Proc. Virtual Worlds and Simulation Conference, San Antonio, Texas},
  month = {Jan},
  type = {mod;bu;cv},
  review = {abs/conf}
}

Downloads: 0