Simulation of movement detection by direction-selective ganglion cells in the rabbit and squirrel retina.

A veto-gate model of movement detection by direction-selective ganglion cells in the vertebrate retina, first proposed by Barlow and Levick (1965), provides the basis for a model described in this study. The model is a simple network consisting basically of (1) two subunits that have receptive fields with a center-surround organization and an adaptational gain control, (2) a lateral inhibitory pathway, (3) a site of nonlinear interaction, followed by (4) a leaky temporal integrator. The model is tested by comparing its basic properties to those reported in the physiological literature on rabbit and squirrel direction-selective retinal ganglion cells. It is shown that the physiological findings on sensitivity to flashes, moving spots or slits, and phi-movement stimuli, can be mimicked quite well by our model. Similarities between the component processes of the subunits and known retinal processes are pointed out. The simulation studies shed a new light on some of the known properties and suggest several new, more revealing, physiological experiments. Such experiments are necessary to develop a full specification of this type of model and to fix more parameter values than is possible at present. Results of some critical experiments are predicted to enable physiologists to falsify or corroborate the model. The simulation studies also help to distinguish use from abuse of this type of model in explanations of psychophysical findings. For example, neither the most complete Barlow-Levick detector nor any stripped-down versions that retain a temporally extended lateral inhibition (which is essential to mimick the physiological findings), respond well to moving random-pixel arrays.