Figure 4. Cortical EN model reproducing differences between iso-orientation domains and pinwheel centres.

(a) Model schematic: inputs were passed through a photoreceptor compressive nonlinearity. The retinal output was convolved with difference-of-Gaussian LGN receptive fields and combined in a push–pull manner, to create elongated cortical receptive fields that were run through an expansive spiking nonlinearity. Neighbouring V1 neurons had orientation preferences corresponding to iso-orientation and pinwheel domains. The cortical components were excitation from surrounding oriented neurons, followed by divisive self-normalization with the output of a non-oriented inhibitory neuron added to the denominator. (b–d) Left: predictions for iso-orientation domain cortical output (red triangles) versus thalamic input for narrowly tuned neurons (black circles). Right: predictions for pinwheel domain cortical output (blue triangles) versus thalamic input for broadly tuned neurons (black circles). (b) Orientation tuning. (c) Contrast response. (d) Responses to images of increasing orientation complexity: preferred grating and plaids of 2, 4 and 8 equally spaced orientations including the preferred. (e) Top row: a natural image with salient edges and surface patterns (Lu, K. Peacock wooing peahen (Creative Commons Attribution 2.0 License), http://birds.wikia.com/wiki/File:Peacock%2Bwooing%2Bpeahen-2008.jpg, 2003). Second row: images processed through the model: (left) two bars and (right) a rosette pattern. Third row: heat maps showing summed response of eight narrowly tuned model neurons with equally spaced orientation preferences in an iso-orientation domain. Bottom row: heat maps showing summed response of eight broadly tuned model neurons with equally spaced orientation preferences in pinwheel centre.