Neural network model for completing occluded contours

— This paper proposes a neural network model capable of completing partly occluded contours. The model is a hierarchical multi-layered network. Using the responses of bend-extracting cells, which are found in area V2 and respond selectively to a particular angle of bend, the model extrapolates the curvature and location of the occluded contours. Keywords— Vision, Occlusion, Amodal completion, Neural network, Bend-extracting cell 1 [1] V1 V2 [2] 2 1 U0 UG US1 S1 V1 US1 US1 V1 UMS1UM UMG U0 UG US1 UC1 US2 US2e W0 WG WC1 WS2 stimulus 1: F A C B α α−β α+β F (a) gate AB C (b) 2: C1 UC1 [3] U0 UM → UMG → UMS1 → US1 [1] UC1 S2 US2 S2 V2 [2] US2e US2 2(a) A S2 S2 B AB p A A p C S2 US2 A B US2e C 

• Type: Journal
• Year: 2010
• Where: NN
• Authors: Kunihiko Fukushima
• Keywords: Neural Network, Neural Network Model, Neural Networks, NN 2010, UMG U0 UG