Modeling spatial filters of primary visual cortex

Here we address the issue of unsupervised learning for primary visual cortex (V₁) formation. According to the most popular model, single cells in the primary visual cortex are spatio-temporal filters. Receptive fields of simple cells can be subdivided into separate excitatory and inhibitory regions, which can be described by a Gaussian-weighted sine wave. These cells transform the fairly broadband input, arriving from lower visual areas, into bandlimited information in the orientation, spatial frequency and temporal frequency domains. The existence of relatively independent spatial frequency and orientation selective channels in the visual system has been shown by human psychophysics. Inhibitory interactions between neighbouring channels were suggested to perform more complex image-analysis, like texture processing or detecting long and smooth contours. Our goal is to introduce a model that imposes minimal assumptions and beyond producing spatial filters can account for some psychophysical results with the help of nonlinearities. Efforts to design self-organizing neurocontrol led to spatial filters. Here, we shall try to identify these filters with the local filters of V₁