Purchase this article with an account.
Thomas Serre, Tomaso Poggio; Standard model v2.0: How visual cortex might learn a universal dictionary of shape components. Journal of Vision 2005;5(8):742. doi: 10.1167/5.8.742.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The tuning properties of neurons in inferotemporal (IT) cortex are likely to play a key role for visual perception in primates and in particular for their object recognition abilities. The tuning of specific neurons probably depends, at least in part, on visual experience.
We describe a model of plasticity and learning in V4 and IT extending the initial version of the standard model of object recognition in cortex [Riesenhuber and Poggio, Nat. Neurosci. 1999] — that accounts for known physiological data. When exposed to many natural images the model generates a large set of shape-tuned units which support robust recognition performance and which can be interpreted as a universal dictionary of shapes with the properties of overcompleteness and non-uniqueness. Preliminary results suggest that the set of shape-tuned units obtained is consistent with recent physiological data collected in V4, see abstract by [Cadieu et al, VSS 2005]. We also show that the model can handle the recognition of different object-categories in natural images at the level of the best existing computer vision recognition systems.
This PDF is available to Subscribers Only