Purchase this article with an account.
Valerio Mante, Matteo Carandini; Energy models and the mapping of multiple features in visual cortex. Journal of Vision 2004;4(8):12. https://doi.org/10.1167/4.8.12.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Optical imaging studies of primary visual cortex (V1) typically compute orientation maps from responses to long drifting bars or gratings. Basole, White & Fitzpatrick (Nature, 2003), however, recently found that these maps are not invariant: if measured with short bars, they depend on direction of motion and speed. The authors suggested that their results are predicted by a simple motion energy model. We tested this hypothesis by simulating V1 responses to drifting bars; model neurons integrate stimulus energy over a small region of three-dimensional frequency space (Watson & Ahumada, JOSA, 1985), determined by their preferences for spatial frequency, temporal frequency, and orientation. We then computed population responses by summing over all neurons preferring the same orientation. The model correctly predicts how population responses depend on bar orientation, direction, length, and speed. In particular, it predicts how bars of the same orientation differing in speed or direction can yield different population responses, while bars differing in orientation (with appropriate speed and direction) can yield similar population responses. The model also trivially predicts the occurrence of motion streaks for fast moving dots (Geisler, Nature, 1999). These and other results are best understood by visualizing stimulus energy and receptive fields in frequency space. Even though an energy model makes no prediction as to the layout of orientation maps on the cortical surface, it correctly predicts population responses. This result provides further evidence that V1 population responses are well described by filters that integrate stimulus energy.
James S. Mc Donnell Foundation, Smith-Kettlewell Eye Research Institute
This PDF is available to Subscribers Only