Abstract
Construction of surface is a crucial step in the integration of image fragments towards shape representation and object recognition. Although neural basis for the surface construction is unknown, border-ownership (BO) selective cells that signal the direction of figure (DOF) appear to play a crucial role. Onset synchronization of BO cells (Dong et al., JoV 2008) is a plausible candidate for binding neural signals to represent surface and its shape. Together with the synchronization, the responses of V1 cells at medial axis (MA; Lee et al., Vis. Res., 1998) that is located at equidistance from nearby contours may play a crucial role in the early stage of surface representation. If the propagation of spikes from BO cells is biased toward DOF, the spikes provide sufficient signals for the reconstruction of surfaces by means of MA even at the presence of occlusion. We investigated computationally what signals emerge, with regard to the surface representation, in a biophysically detailed model of V1–V2 network that includes V1 cells, BO cells, lateral connections, and bidirectional connections between V1–V2. The model reproduced the onset synchronization of BO cells. We carried out the simulations of the model with a variety of shapes including those of natural objects to examine how the spikes of BO cells at the synchronization propagate in the network. V1 cells located at equidistance from the contours resulted in strong responses, generating the representation of MA. The correlation coefficient between the MA represented by the model responses and that computed from theory was 0.7, and the mean error of shape reconstruction with respect to the original shape was 0.3, indicating good representation of MA. These results suggest a crucial role of the onset synchronization of BO cells for the early representation of surface and its shape by means of MA.