Abstract
The majority of cortical neurons in area V1 are complex cells with highly nonlinear response properties. While many properties of these cells have been studied in the past, complete descriptions of their stimulus response functions remain difficult to obtain. Here, we recorded from cat primary visual cortex and obtained second-order Wiener kernels using a random-bar white-noise stimulus. For each cell, singular value decomposition of the second-order Wiener kernel identified a set of eigenvectors whose receptive fields had separate ON and OFF sub-regions, resembling the linear receptive fields of simple cells. We found that predictions based on these eigenvectors provide a more accurate description of the cell's response than predictions based on its linear receptive field. In addition we tested complex cells by creating stimuli that were orthogonal to the cell's eigenvectors. Many of these cells were not significantly modulated by this stimulus, indicating that these eigenvectors accounted for a large fraction of the neuron's response properties. These results support the notion that the response properties of complex cells can be approximated as the nonlinear combination of simple-cell-like functional subunits.