Cortical Surround Interactions and Perceptual Salience via Natural Scene Statistics
Figure 7
Expansion of receptive field size at low contrast.
(A) Normalized mean response rate of a V1 neuron (left; [8]) and the model (right) as a function of stimulus size at two contrasts. The largest size we used to test the model covers the full extent of the surround RFs, and larger sizes will not change RFs outputs; for the 5 largest sizes tested, we observed little change in surround RFs outputs and model responses. Stimulus orientation and spatial frequency are optimal for the neuron. The insets show some example stimuli. The arrows (gray, low contrast; black, high contrast) indicate peak response diameter - or RF size. (B) Probability that center and vertical surround RFs are co-assigned to the same normalization pool, and therefore contribute to the divisive normalization of the model response; the stimuli are the same as in (A). At the smallest non-zero size, surround RFs are silent and therefore the surround is not co-assigned. At intermediate sizes, surround RFs outputs are weaker than center RFs outputs and co-assignment probability increases with contrast, as illustrated in Fig. 5. At the largest sizes, RFs outputs are similar between center and surround and the surround is co-assigned.