Visual Scene Segmentation: The Local/Global puzzle

Nava Rubin

Monday, February 24, 2003, 02:00pm - 03:00pm

New York University, Center for Neural Science


Early in visual processing the retinal image is highly fragmented, with
each visual neuron having a very small (local) receptive field.� How is
this information 'put back together' to represent the global objects we
consciously perceive?� A common view is that this is done by a
progression of feed-forward connections to higher-order visual areas,
where cells have large receptive fields.� But in the domain of
segmentation -- the process of parsing the retinal image into the
different surfaces comprising the scene -- this view presents serious
challenges.� I will summarize results from psychophysics, modeling and
brain imaging (fMRI) studies that bear on the issue, and propose a
theoretical framework to account for existing data.� In this framework,
feedback from higher-order to early visual cortical areas continuously
plays a role in segmentation.

Nava Rubin