Neural dynamics in a recurrent network model of primary visual cortex

The neural network of the primary visual cortex (V1) is an example of a recurrent network with translation invariant neural connections. The recurrent interactions make the cortical outputs a complex nonlinear function of visual inputs in the service of a very difficult computational task: pre-attentive visual segmentation. Understanding the nonlinear dynamics of the neural circuit is a key to appreciating the computational potential of the primary visual cortex, which is an early stage in the visual pathway, and is usually thought of as a low level visual area. This paper describes an analytical study of the recurrent neural dynamics in the first biologically based model of V1 to achieve simultaneously the computations of region segmentation, figure-ground segregation, and contour enhancement. By relating neural connections to the network behavior, our analysis enables the educated design of network models for classes of computation. Many of the analytical techniques can be applied to other recurrent networks with translation symmetry in the connections.