Abstract

A new mathematical description of visual receptive fields in the striate cortex is proposed. The hypothesis is advanced that receptive field line­weighting profiles along the axis of maximum response are describable as Gaussian derivative functions. Such functions are a close approximation to the optimum functions theoretically possible for the detection and precise localization of visual stimuli in the presence of noise. The shapes of the spatial frequency sensitivity tuning curves of fifty-five monkey cortical cells were well characterized by a single Gaussian directional derivative, most often the first through fourth derivative, with corresponding half-amplitude bandwidths in the range of 2.6-1.2 octaves. The receptive field shapes of cat and monkey cells in the spatial domain, as measured, with bar stimuli, were also well described by a single Gaussian derivative term. Gaussian derivatives generally described, better than Gabor functions, (1) the locations of the zero crossings of receptive field profiles in the spatial domain; (2) the shape of cell’s spatial frequency tuning curve, at the low and high frequency extremes; and (3) the symmetry properties of fields in the spatial domain as a function of their independently measured bandwidths. A new neural model for the formation of receptive fields described how simple differences of offset Gaussians (DOOGs) with identical standard deviations, such as are found in the monkey X-cell pathway, can easily give rise to Gaussian derivativelike functions of any order.

© 1985 Optical Society of America