Continuum model of primary visual cortex for contour detection

Summary

Date: 
February 20, 2015 - 11:00am
Location: 
NW 425
About the Speaker
Name: 
Vijay Singh (Nemenman Lab, Emory University)

Information processing on biological networks takes places at many scales. Coarse graining these networks, one can identify the features that are essential for such information processing. I will talk about my work related to coarse-graining networks to identify such features. In particular, I will talk about a coarse grained description of the neurons in primary visual cortex (V1) to study contour detection in V1. In a cluttered visual scene, it becomes very difficult to detect contours when parts of the object are occluded. To identify the features essential for contour detection, we model the neural dynamics in V1 in terms of a continuous director field that describes the average rate and the average orientation preference of active neurons at a particular point in the cortex. We then use a linear-nonlinear dynamical model with long range co-circular interactions to enforce long-range statistical context present in the analyzed visual scene. The resulting model has substantially fewer degrees of freedom than traditional models, and yet it can distinguish large contiguous objects from the background clutter by suppressing the clutter and by filling-in occluded elements of object contours. This results in high-precision, high-recall detection of large objects in cluttered scenes.