Fig. 5
From: Overcoming the limitations of motion sensor models by considering dendritic computations
The MS-INRF model reproduces classical findings about motion masking. (a, b) The MS-INRF response to a signal consisting of a moving sinewave grating (orange) is greatly reduced when we add to the signal jittering noise that has a similar spatial frequency (response to signal plus noise in blue); responses are averaged across different presentations with varying spatial phase of the stimuli. (a) Simulation of retinal result, where the MS-INRF model is applied directly to the stimulus. (b) Simulation of cortical result, where the stimulus is filtered with a Difference of Gaussians (DoG) kernel in order to emulate processing in the lateral geniculate nucleus (LGN), and then passed to the MS-INRF model. (c) Reciprocal of the data shown in (b) (normalized), which may be interpreted as a reduction in sensitivity of the MS-INRF response. The narrow tuning of the masking effect to the spatial frequency of the noise is in agreement with neurophysiological results of simple and complex cell responses (d, e adapted from70, their Figs. 3 and 4 respectively) and with psychophysical results observed in humans (f, adapted from50, their Fig. 1). The phenomenon where the tuning to spatial frequency is narrowed as we add processing stages is consistent with classical physiological results, as retinal ganglion cells are tuned to a broad range of spatial frequencies, LGN cells to a narrower range, and simple V1 cells to an even narrower range of spatial frequencies70.
