Receptive fields of visual cortical neurons as derived by infomax

Complex cells, that is, visual neurons in the brain, were investigated from an information-theoretic viewpoint. Responses of a complex cell to visual stimuli are well described by a phenomenological model called the energy model. The energy model describes the output of a complex cell as a squared sum of the outputs of two linear operators. An information-maximization problem was solved in order to determine the two linear operators. To obtain spatially localized operators (receptive fields), we assumed a localization term in the objective function. At a low signal-to-noise-ratio limit, two linear operators characterized by a quadrature pair of Gabor functions were obtained as solutions to the maximization problem. This result agrees with the energy model; thus, it strongly suggests that complex cells are optimally designed from an information-theoretic viewpoint. © 2001 Scripta Technica.