Abstract
A long standing proposition in psychophysics is that there are two temporal channels in spatial vision: One temporal lowpass filter which is spatially bandpass and one temporal bandpass filter which is spatially lowpass. Here we equate these two channels with the midget and parasol ganglion cells of the primate retina respectively. Parasol cells show a frequency doubling response at high spatial frequencies, i.e. they respond to both polarities of high spatial frequency gratings. This is usually thought to result from them pooling over smaller units. If true, this explanation predicts that signals with both high spatial and high temporal frequency should be detectable, but their spatial orientation and phase should not be perceivable, i.e. parasol cells act not as labelled lines in this scenario. We confirm this prediction by finding a difference between detection and orientation discrimination thresholds at high spatial and temporal frequency, not observed at other spatial-temporal frequency combinations. We model midget and parasol cells using a standard divisive model dividing a generalized Gaussian center by a larger surround. When scaling cells proportional to their known separation in the retina, we can make predictions for the perception performance of human observers assuming optimal decoding after a single quadratic non-linearity. To confirm the predictions of our model we measured contrast sensitivity functions (CSFs) under diverse conditions and fit data from the literature. Our model fits the CSF under different temporal conditions and changes in size with fixed spatial parameters and can thus replace previous CSF models which require new parameters for each condition and have no mechanistic interpretation. Finally, our model provides a counter hypothesis to the textbook explanation for the CSF which describes it as the envelope of the spatial frequency tuned channels in V1; instead, we believe its shape to result from properties of retinal cells.