J. Anthony Movshon

17 janvier 2013

Center for Neural Science, New York University, New York, USA

Cortical tasseography

Invité par Frédéric Chavane

Abstract : Extrastriate visual area V2 is the largest and in some ways the most enigmatic cortical visual area. V2 receives a strong direct input from V1, and depends on the functional integrity of V1 for its visual responsiveness. V2 is often assumed to execute the first steps toward the representation of visual scenes and forms evident further downstream in the extrastriate pathways, but there is no satisfactory account of the nature of those steps. In fact, by most measures cells in V2 respond similarly to those in V1. We have attempted to account for the response properties of V2 neurons by identifying and characterizing the V1 neurons that project to V2, and then using these properties as the basis for a feedforward model similar to one we used in previous work to account for the responses of cells in MT. This model accounts for a respectable fraction of the explainable variance of V2 responses, but is not easy to summarize because it is not founded on an explicit theory of visual processing. We therefore turned to a family of stimuli based on a model of natural image structure. Many V2 cells respond more vigorously to synthetic “naturalistic” texture stimuli than to matched “noise” stimuli ; V1 cells rarely differentiate these stimuli. To relate these differential responses to perception, we asked human observers to discriminate between images drawn from the two ensembles. Strikingly, the texture categories for which perceptual discrimination is best are those that exhibit the largest differential responses, both in macaque V2 neurons and human fMRI. We conclude that neuronal responses in V2, but not V1, are sensitive to the joint statistics of local structure found in natural images, and may reflect an emerging specialization for these features in the extrastriate visual pathway.

CNRS logo université Aix Marseille logo | plan du site | mentions légales | contact | admin | intranet | intcloud |