Modelling the relative disparity selectivity of V2 neurons

Psychophysical studies suggest that stereopsis relies primarily on relative disparities, yet V1 neurons are selective for absolute disparity (Cumming and Parker, 1999; J. Neurosci. 19(1981-2088)). We have found that a proportion of V2 neurons in the awake, behaving monkey show the novel property of being sensitive to relative, not absolute, disparity (Thomas, et al. 1999; Soc. Neurosci. Abs. 25(2060)). Here, we implement a model of relative disparity selectivity by combining the responses of absolute-disparity sensitive elements, similar to real V1 neurons. The model is selective to the relative disparity between two spatially segregated regions - the model's receptive field 'centre' and 'surround' (which is also the arrangement of the experimental stimulus). The model's construction is analogous to the disparity-energy model (Ozhawa et al. 1990; Science, 249 (1037-1041)) but instead of monocular filters, we substitute absolute-disparity sensitive neurons. We combine pairs of such neurons (one from the centre, one from the surround) by summing their responses followed, crucially, by an output non-linearity (half-squaring). The difference between a pair's absolute disparity tuning curves defines a relative disparity between centre and surround; neurons are paired such that this relative disparity is consistent across pairs. The model V2 neuron is the sum of all such pairs' responses: this displays a constant relative disparity selectivity, independent of the absolute-disparity. Thus the same cortical computations and connectivity that are capable of generating selectivity for absolute disparity in V1 may also be responsible for the emergence of relative disparity sensitivity in V2.