Curvature and orientation estimation by neuronal structures

The paper presents a simple model of curvature and orientation estimation by neural networks where a pair of neurons is used to approximate the partial differential operators needed for curvature and orientation estimation. The influence of neuronal morphometry in the estimation of curvature and orientation is investigated and discussed. In addition, the biological plausibility of the model is discussed, and simulation results are presented along a sequence of increasing plausibility and sophistication. Steerable filters are considered as a means to increase the model efficiency.