The origins of spatial frequency tuning in macaque visual cortex

P G Levy, C M Ziemba, J A Movshon, E P Simoncelli and R L T Goris

Published in Annual Meeting, Neuroscience, Nov 2016.
Neurons in primary visual cortex (V1) are selective for the spatial frequency of image features. This selectivity first arises in the retina, is modified by the LGN, and is further sculpted by cortical mechanisms. To identify the contributions of these elements, we measured responses of V1 neurons in anesthetized macaques to mixtures of sinusoidal gratings dispersed in spatial frequency. For a fixed total contrast, we varied the frequency dispersion and studied selectivity for the modal frequency of the mixtures. Many neurons exhibited contrast-dependent frequency tuning: At low contrast, neurons had narrower octave bandwidths and preferred lower spatial frequencies. These contrast dependencies were more pronounced for responses to more dispersed stimuli.

To explain these results, we developed a cascaded model of cortical computation. In the first stage of the model, visual stimuli are processed in parallel by two linear-nonlinear channels, each containing the half-wave rectified responses of center-surround filters whose properties are based on those of parvocellular and magnocellular LGN cells that project to cortex. In the second stage, these channel responses are combined linearly, scaled by a contrast gain control, and passed through a nonlinearity to obtain a firing rate. The model simulates cortical responses to mixture stimuli well, and suggests that contrast-dependent spatial frequency tuning might emerge from combinations of inputs with different contrast sensitivity and spatial frequency selectivity.

  • Listing of all publications