Abstract
I have proposed a model of how the primate visual system extracts a speed signal from the activity of a small set of Middle Temporal (MT/V5) neurons (Perrone, JOV, 2012). I suggest that the visual system uses the activity from a number of velocity channels each made up of a 'triad' of MT neurons tuned to 2V, V, .5V and spatial frequencies (u/2, u, u). A unique aspect of the model is that the .5V unit is a component type neuron whereas the other two are pattern types. For a compound, horizontally moving stimulus made up of two sine wave gratings (sf1 = sf2 = u) with speeds V and .5V, the model speed output is given by the weighted vector average of the MT triad activity (0 x 2V, 1 x V, 1 x .5V) = .75V′. The 2V unit is inactive because it is tuned to speed 2V and sf = u/2. If we now replace the .5V grating with a 150° plaid (overall speed = .5V) the activity distribution is (0 x 2V, 1 x V, 0 x 5V) with predicted perceived speed = V′ because the plaid does not activate the .5V MT component unit. The ratio of the perceived speeds (Rpg) for the two types of compound stimuli (with plaid/with grating) should therefore be 1.33 (1/.75). This prediction was tested using a psychophysical procedure with the speed of the two types of compound stimuli compared against a single variable grating. The mean Rpg value from 3 observers was 1.32 (95% CI = 1.28 – 1.37). This supports the idea that speed estimation involves an MT component neuron and challenges velocity population code models based on a winner-takes-all mechanism or models that do not distinguish between a .5V grating and plaid (both types predict Rpg = 1.0).
Meeting abstract presented at VSS 2016