Abstract
We investigated the extent to which the transformations performed by V1 neurons are consistent with an LN model of spatial filtering by examining their responses to two-dimensional Hermite functions (TDH's). TDH's are spatial patterns that form a complete basis set and are simultaneously localized in space and spatial frequency. They also form a natural hierarchy: the rank 0 function is a Gaussian; functions of higher rank are progressively broader in space and spatial frequency. Unlike Gabor functions, the higher-rank functions are intrinsically two-dimensional. Within each rank, the functions can be organized into Cartesian and polar patterns. Cartesian and polar patterns are precisely matched in spatial extent and spectral content but differ in their two-dimensional structure. 21 of 31 cells in cat V1 responded robustly to transient presentation of Cartesian and polar TDH functions of ranks 0 through 7. Responses to each set of functions yielded independent estimates of spatial filtering properties and the degree of nonlinearity. For less than half of the neurons, the characterizations obtained from the two basis sets were similar in these respects. Eight neurons showed differences in their spatial filtering properties, including a clear shift in orientation tuning in some cases. Three neurons behaved more non-linearly in response to Cartesian stimuli than to polar stimuli. Two neurons, both infragranular, responded only to the Cartesian stimuli. These behaviors are inconsistent with the notion that V1 neurons act as oriented filters followed by a simple nonlinearity, since the two-dimensional nature of the stimuli alters qualitative aspects of the response. We speculate that the above phenomena reflect an interplay between simple feedforward filtering and spatial processing intrinsic to V1.