|
Institute for
|
|
|
Julian Martin Fernandez
Object Recognition
|
We are trying to model how the visual
system learns to represent objects hierarchically. This is a promising area
of research because most current models of object recognition are view-based,
rather than parts-based, and the few existing parts-based models are often
too simple to deal with the issue of how an object’s hierarchical
structure is represented. I think it is necessary for models of
visual object representation to take a hybrid approach, in which the two
methods, view-based and parts-based, both contribute. I have been
concentrating in developing a sophisticated parts-based model that is able to
hierarchically represent objects and works with the help of existing
view-based models. If, when fully implemented, the model works as expected,
it will explain a range of data on human learning of object and scene
structure significantly beyond the range explained by current models. This work is still at an exploratory
stage. More details coming soon. |