A model visual cortex incorporating intrinsic horizontal neuronal connections

Abstract The model proposed here is based on a localized spatial frequency analysis model for the primary visual cortex (i.e., the Gabor function model for simple cells and the Fourier amplitude model for complex cells). In this model the statistical properties of the signal and noise components are stored in the intrinsic horizontal connectivity pattern, which is used aspriori knowledge to extract the signal from noisy input. The method of the Wiener filter, or the least-squares filter, is applied to the activity pattern of the model complex cells, and the optimal horizontal connectivity pattern for estimating the signal is derived. As a simple but important example, the model horizontal connections required to extract oriented features such as lines or edges are calculated. The results of numerical simulations show that the model horizontal connections can, as expected, extract the local features from degraded input. The obtained model horizontal connectivity is, in accord with experimental observations by Ts'o, Gilbert & Wiesel [Journal of Neuroscience, 6 , 1160 (1986) ], highly specific in connecting cells having similar orientation selectivity.